ORDER GRANTING IN PART AND DENYING IN PART PERFECT 10’S MOTION FOR PRELIMINARY INJUNCTION AGAINST GOOGLE

I. INTRODUCTION

The principal two-part issue in this case arises out of the increasingly recurring conflict between intellectual property rights on the one hand and the dazzling capacity of internet technology to assemble, organize, store, access, and display intellectual property “content” on the other hand. That issue, in a nutshell, is: does a search engine infringe copyrighted images when it displays them on an “image search” function in the form of “thumbnails” but not infringe when, through inline linking, it displays copyrighted images served by another website?

Plaintiff Perfect 10, Inc. (“P10”) filed separate suits against Google, Inc. and against Amazon.com, Inc. and its subsidiary, A9.com, Inc. (collectively, “Amazon”), alleging copyright and trademark infringement and various related claims. The suits were consolidated. P10 moves now for a preliminary injunction against both Defendants, solely on the basis of its copyright claims. P10 seeks to prevent Defendants’ image search engines from displaying “thumbnail” copies of PlO’s copyrighted images and also from linking to third-party websites which host and serve infringing full-size images.

The Court conducted a hearing on November 7, 2005. The Court now concludes that Google’s creation and public display of “thumbnails” likely do directly infringe PlO’s copyrights. The Court also concludes, however, that P10 is not likely to succeed on its vicarious and contributory liability theories.

This Order will address PlO’s motion for preliminary injunctive relief against Google. Amazon licenses from Google much of the technology whose use by Amazon P10 challenges. A separate order will address PlO’s motion against Amazon.

II. BACKGROUND

A. The Parties

1. Perfect 10

P10 publishes the adult magazine “PERFECT 10” and operates the subscription website, “perfectlO.com,” both of which feature high-quality, nude photographs of “natural” models. Pl.’s Zada Decl. ¶¶ 9-10. During the last nine years, PIO has invested $36 million to develop its brand in its magazine and its website. Id. ¶ 11. This investment includes approximately $12 million spent to photograph over 800 models and create 2,700 high quality images that have appeared in its magazine, along with an additional approximately 3,300 images that have appeared on per-fectl0.com. Id. P10 has obtained registered copyrights for its photographs from the United States Copyright Office. Id., Ex. 1.

P10 generates virtually all of its revenue from the sale of copyrighted works: (1) it sells magazines at newsstands ($7.99 per issue) and via subscription; (2) it sells website subscriptions to perfectl0.com for $25.50 per month, which allow subscribers to view P10 images in the exclusive “members’ area” of the site ; and (3) since early 2005, when P10 entered into a licensing agreement with Fonestarz Media Limited, a United Kingdom company, for the worldwide sale and distribution of P10 reduced-size copyrighted images for download and use on cell phones, it has sold, on average, approximately 6,000 images per month in the United Kingdom. Id. ¶ 16. Aside from the licensing agreement with Fones-tarz Media Limited, P10 has not authorized any third-party individual or website to copy, display, or distribute any of the copyrighted images which P10 has created. Id. ¶ 17.

2. Google

Google describes itself as a “software, technology, Internet, advertising, and media company all rolled into one.” Google, Inc., 2004 Annual Report (Form 10-K), at 9. Google is one of the most highly frequented websites on the internet. Pl.’s Zada Deck, Ex. 3 (report from Alexa Internet, Inc. showing http://www.goo-gle.com/ranking as the third most visited site in the world). Google operates a search engine located at the domain name “google.com.” Google’s search engine indexes websites on the internet via a web “crawler,” ie., software that automatically scans and stores the content of each website into an easily-searchable catalog. Def.’s Levine Decl. ¶¶ 13-14. Websites that do not wish to be indexed, or that wish to have only certain content indexed, can do so by signaling to Google’s web crawler those parts that are “off limits.” Google’s web crawler honors those signals.

Google operates different search engines for various types of web content. All search queries are text-based, ie., users input text search strings representing their query, but results can be in the form of text, images, or even video. Id. ¶ 21. Thus, for example, Google’s basic web search, called Google Web Search, located at http://www.google.com/, receives a text search string and returns a list of textual results relevant to that query. Google Image search, on the other hand, receives a text search string and returns a number of reduced-sized, or “thumbnail” images organized into a grid.

Google stores content scanned by its web crawler in Google’s “cache.” For Google Web Search, because its “web page index is based entirely on the textual part of web pages and not the images, [its] web page cache contains only the text pages, and not the images that those pages include when displayed.” Def.’s Levine Decl. ¶ 21. For Google Image Search, too, the results depend solely on the text surrounding an image. Id. ¶ 22. But for Image Search, Google also stores thumbnails in its cache, in order to present the results of the user’s query. Def.’s MacGil-livray Decl. ¶ 3 (“The browser obtains ‘thumbnail’ images from Google’s server....”); Pl.’s Mausner Decl., Ex. 118, Google’s Resp. to PlO’s Req. for Admis. No. 24 (“Google admits that its servers store reduced-size extracts of images.”). A user of Google Image Search can quickly scan the grid of returned thumbnails to determine whether any of the images responds to his search query. He “can then choose to click on the image thumbnail and show more information about the image and cause the user’s browser ... to open a ‘window’ on the screen that will display the underlying Web page in a process called ‘framing.’ ” Def.’s MacGillivray Decl. ¶ 3.

“Framing” is a method of “combining] multiple pages in a single window so that different content can be viewed simultaneously, typically so that one ‘frame’ can be used to annotate the other content or to maintain a link with an earlier web page.” Def.’s Levine Decl. ¶ 24 n. 1. In other words, when a user clicks on a thumbnail returned as the result of á Google Image Search, his computer pulls up a page comprised of two distinct frames, one hosted by Google and a second hosted by the underlying website that originally hosted the full-size image. The two frames are divided by a gray horizontal line a few pixels high. The upper frame is the Google frame. It contains the thumbnail, retrieved from Google’s cache, and information about the larger image, including the original resolution of the image and the specific URL associated with that image. The Google frame also states that the thumbnail “may be scaled down and subject to copyright” and makes clear that the upper frame is not the original context in which the full-size image was found, stating, “Below is the image in its original context on the page: http://.” The lower frame contains, or shows, the original web page on which the original image was found. Google neither stores nor serves any of the content (either text or images) displayed in the lower frame; rather, the underlying third party website stores and serves that content. Id. ¶¶ 27-29. However, because it is Google’s web-page that composites the two frames, the URL displayed in the browser’s address bar displays “images.google.com.” Id.

Attached hereto as Exhibit A is an example of the two-frame structure just described, containing in the upper frame one of the thumbnail images that appeared on the display of thumbnails retrieved by an image search for “Vibe Sorenson,” a P10 model.

Google generates much of its revenue through two advertising programs: Ad-Words, for advertisers, and AdSense, for web publishers. Def.’s MacGillivray Decl. ¶ 9. Through AdWords, advertisers purchase advertising placement on Google’s pages, including on search results pages and Google’s Gmail web-based email service. Id. Google’s AdSense program allows pages on third party sites “to carry Google-sponsored advertising and share [with Google the] revenue that flows from the advertising displays and click-throughs.” Id. ¶ 10. “To participate [in AdSense], a website publisher places code on its site that asks Google’s server to algorithmically select relevant advertisements” based on the content of that site. Id.

B. Procedural History

On November 19, 2004, P10 filed suit against Google asserting various copyright and trademark infringement claims: (1) direct copyright infringement, (2) vicarious copyright infringement, (3) contributory copyright infringement, (4) circumvention of copyright protection systems under the Digital Millennium Copyright Act (“DMCA”), (5) direct trademark infringement, (6) contributory trademark infringement, (7) vicarious trademark infringement, (8) trademark dilution (federal), (9) unfair competition, (10) wrongful use of a registered mark, (11) trademark dilution (state), and (12) violation of rights of publicity. Compl. ¶¶ 35-115.

C. Proposed Injunctive Relief

P10 seeks to preliminarily enjoin Google from engaging in the following activities:

PL’s Proposed Prelim. Inj. ¶ 1.

III. DISCUSSION

A. Legal Standard for Preliminary Injunction

1. General Principles

“A preliminary injunction should be granted if a plaintiff can show either: (1) a combination of probable success on the merits and the possibility of irreparable harm; or (2) that serious questions are raised and the balance of hardships tilt in the plaintiffs favor.” Elvis Presley Enters., Inc. v. Passport Video, 349 F.3d 622, 627 (9th Cir.2003). In any preliminary injunction analysis, courts also look to “whether the public interest will be advanced by granting preliminary relief.” Preminger v. Principi, 422 F.3d 815, 823 (9th Cir.2005).

2. Affirmative Defenses

Google does not contest that photographs are copyrightable subject matter or that PlO’s certificates of copyright registration. have sufficiently established its ownership. Google does, however, dispute PlO’s contention that its copyright interests have been directly infringed. Although Google admits creating and storing thumbnail copies of PlO’s full-size images (found on third-party websites), as well as displaying those thumbnails as search results on Google Image Search, it argues that such use is protected under the fair use doctrine, as codified by 17 U.S.C. § 107 .

“The plaintiffs burden of showing a likelihood of success on the merits includes the burden of showing a likelihood that it would prevail against any affirmative defenses raised by the defendant.” Dr. Seuss Enters., L.P. v. Penguin Books USA, Inc., 924 F.Supp. 1559, 1562 (S.D.Cal.1996) (citing Atari Games Corp. v. Nintendo, 975 F.2d 832, 837 (Fed.Cir.1992)), aff 'd, 109 F.3d 1394 (9th Cir.1997); accord Religious Tech. Ctr. v. Netcom Online Communication Servs., 923 F.Supp. 1231 , 1242 n. 12 (N.D.Cal.1995) (“Even though fair use is an affirmative defense[,] plaintiffs, as the parties moving for a preliminary injunction, have the burden of proving a likelihood of success on their infringement claim, including the fair use defense.”); 2 William W. Schwarzer, et al., California Practice Guide, Federal Civil Procedure Before Trial ¶ 13:47 (2000); Vision-Ease Lens, Inc. v. Essilor Int’l SA, 322 F.Supp.2d 991 , 995 n. 5 (D.Minn.2004) (citing Dr. Seuss).

P10 points out that the Ninth Circuit, in affirming the district court’s ruling in Dr. Seuss, stated that because “fair use is an affirmative defense, [defendant] must bring forward favorable evidence” relevant to the fourth fair use factor. Dr. Seuss, 109 F.3d at 1403 . However, the defendant had presented no evidence whatsoever on that factor. The Ninth Circuit stated that the absence of any evidence made it “impossible to deal with the fourth factor except by recognizing that a silent record on an important factor bearing on fair use disentitle^] the proponent of the defense ... to relief from the preliminary injunction.” Id. (internal quotations omitted) (alteration in original). The Ninth Circuit did not state or indicate that the district court erred in interpreting Plaintiffs burden of proving “likelihood of success on the merits” to include the “likelihood that [plaintiffs] would prevail against any affirmative defenses raised by the defendant.” That the Ninth Circuit required the defendants to submit some quantum of evidence does not mean it had placed the burden on defendant to show that it was entitled to the affirmative defense of fair use.

P10 also cites A & M Records, Inc. v. Napster, Inc., 114 F.Supp.2d 896 (N.D.Cal.2000) [hereinafter “Napster /”], which stated that “defendant bears the burden of proving ... affirmative defenses” on a motion for preliminary injunction. Id. at 912 (citing Bateman v. Mnemonics, Inc., 79 F.3d 1532 , 1542 n. 22 (11th Cir.1996)). On appeal, the Ninth Circuit did not conclusively resolve the issue. It merely stated, “[E]ven if plaintiffs bear the burden of establishing that they would likely prevail against Napster’s affirmative defenses at the preliminary injunction stage, the record supports the district court’s conclusion that Napster users do not engage in fan-use of the copyrighted materials.” A & M Records, Inc. v. Napster, Inc., 239 F.3d 1004 , 1015 n. 3 (9th Cir.2001) (emphasis added). This qualified language does not necessarily suggest that the defendant bears the burden of proving the affirmative defense at the preliminary injunction phase. Indeed, the Ninth Circuit’s analysis proceeded under the assumption that plaintiffs shouldered the burden. Furthermore, the Eleventh Circuit Bateman decision upon which the district court relied did not involve a preliminary injunction.

Accordingly, the Court will follow Dr. Seuss’s conclusion that P10, on its motion for preliminary injunction against Google, carries the burden of overcoming Google’s fair use defense.

3. Prohibitory v. Mandatory Injunction

Google tries to characterize the relief P10 seeks as “mandatory” and points out that “ ‘mandatory preliminary relief is subject to heightened scrutiny and should not be issued unless the facts and law clearly favor the moving party.” Dahl v. HEM Pharm. Corp., 7 F.3d 1399 , 1403 (9th Cir.1993); see also Stanley v. Univ. of Southern California, 13 F.3d 1313, 1320 (9th Cir.1994). P10 contends, however, that “an injunction that requires a defendant to refrain from performing present and continuing acts causing injury is prohibitory,” not mandatory. See First Union Nat. Bank v. Burke, 48 F.Supp.2d 132, 142-143 (D.Conn.1999). “The distinction between mandatory and prohibitory injunctions is not without ambiguities.... ” Tom Doherty Assocs., Inc. v. Saban Entm’t, Inc., 60 F.3d 27, 34 (2d Cir.1995). “[Mjany prohibitory injunctions can easily be restated in a manner that makes them appear mandatory in effect.” First Union Nat. Bank, 48 F.Supp.2d at 143 .

The Court agrees with P10. The proposed injunction would be essentially prohibitory in nature, because it would require Google to cease its allegedly infringing activities. Whatever active steps Google might have to undertake would merely be the means of discontinuing acts of infringement.

B. Likelihood of Success

P10 asserts that Google is both directly and secondarily liable for copyright infringement. P10 alleges that Google’s image search engine directly infringes by copying, distributing, and displaying thumbnails and full-size images of PlO’s copyrighted photographs. P10 alleges that Google is secondarily hable for the actions of third-party websites that host infringing images and unauthorized per-fectl0.com username/password combinations to which Google’s search engine links, as well for the actions of individuals who are led by Google Image Search to infringing images and subsequently download infringing copies themselves.

Google raises several defenses. First, in response to PlO’s direct infringement claims, it argues that (1) many of its actions do not infringe upon any of the exclusive rights granted to the owner of a copyright, and (2) to the extent that its actions do implicate those rights, such use is fair under 17 U.S.C. § 107 . Second, in response to PlO’s secondary liability claims, Google contends that (1) it has not eontrib-utorily or vicariously infringed; (2) it is immune from contributory liability under Sony Corp. of America v. Universal City Studios, Inc., 464 U.S. 417 , 104 S.Ct. 774 , 78 L.Ed.2d 574 (1984) [hereinafter “Sony”]; and (3) it qualifies for protection under the various safe harbor provisions of the DMCA, 17 U.S.C. § 512 (a)-(d).

1. Direct Infringement

To establish direct copyright infringement, a plaintiff must prove two elements: (1) ownership of a valid copyright, and (2) violation of one of the exclusive rights granted under copyright. Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., Inc., 499 U.S. 340, 361 , 111 S.Ct. 1282 , 113 L.Ed.2d 358 (1991).

a. What Actions by Google Allegedly Constitute Direct Infringement?

Although 17 U.S.C. § 106 sets forth six exclusive rights of a copyright holder, the rights in question here are the right to display publicly, the distribution right, and the reproduction right. P10 alleges that Google directly infringes in that it both (1) displays and distributes full-size images hosted by third-party websites, and (2) creates, displays, and distributes thumbnails of PlO’s copyrighted full-size images. P10 contends that these displays and distributions of copyrighted material extend to cell phones as well as computers. Google concedes that it creates and displays thumbnails; it denies that it “displays,” creates, or distributes what is depicted in the lower frame; and it challenges PlO’s argument that any of its activities can be the basis for direct infringement.

The Court will address PlO’s contentions about framed, full-size images first.

b. As to “In-Line Linking,” What Constitutes a “Display”?

There is no dispute that Google “in-line links” to and/or “frames” content that, in fact, is stored on and served by other websites. Whether that conduct constitutes a “display” for purposes of copyright law is the issue.

The terms “link” and “in-line link” can be used in two distinct, but related, ways. “Link” is most commonly used to refer to text or image “hyperlinks” that are displayed on a webpage and that when clicked by the user, transport him to a new page. “In-line link” refers to the process whereby a webpage can incorporate by reference (and arguably cause to be displayed) content stored on another website. Kelly v. Arriba Soft Corp., 336 F.3d 811, 816 (9th Cir.2003) [hereinafter “Kelly II”] (“The in-line link instructs the user’s browser to retrieve the linked-to image from the source website and display it on the user’s screen, but does so without leaving the linking document.”).

There are at least two approaches to defining “display” in the context of in-line linking: what the Court will call (1) a “server” test and (2) an “incorporation” test. The differences in the “server” and “incorporation” tests can be illustrated if we return to the example of SoccerMA-NIA.com. See supra note 7. That fictitious website might contain a single webpage with the text, “We proudly show this photo,” below which appears a photo of the legendary soccer great Pelé that he (Pelé) copyrighted. The mere fact that the Pelé photo appears on SoccerMANIA’s web-page does not necessarily mean that the photo (or even a copy of it) is stored on or transferred via SoccerMANIA.com. Using standard HTML, Soccer MAN IAxom’s webpage might, in fact, be in-line linking to the Pelé photo stored on, say, “Soccer-PASSION.com.” If that is the case, when the person seeking to visit SoccerMA-NIAcom’s webpage uses his browser, the browser would (1) download SoccerMA-NIAxom’s webpage, (2) parse through the various HTML commands of that web-page, (3) per HTML code, display the text “We proudly show this photo,” (4) also per HTML code, follow an in-line link to the image stored on SoccerPASSION.com, (5) download the photo to the user’s computer directly from SoccerPASSION.com, and (6) display the image in the browser below the text. Because the visitor cannot see any of these actions take place, he probably—but mistakenly—will assume that the copyrighted photo of Pelé is stored on and served by SoccerMANIA.com. Indeed, even though the image was actually transferred directly from SoccerPASSION.com, the address shown on the user’s browser will still indicate something akin to “http://www.SoccerMANIA.com/web-page.html”. This is because browsers display the address of the file (here, a web-page) that they are currently rendering; they do not in any way indicate the location from which each component element of a webpage (such as an image) originates.

The question, then, is whether Soecer-MANIA.com, SoccerPASSION.com, or both have “displayed” the copyrighted Pelé photo.

i. The Server Test Embraced by Google

From a technological perspective, one could define “display” as the act of serving content over the web—ie., physically sending ones and zeroes over the internet to the user’s browser. Adopting this definition, as Google urges the Court to do, SoccerPASSION.com would be the entity that “displays” the Pelé image, and Soccer-MANIA.com would not risk liability for direct infringement (regardless of whether its in-line linking would otherwise qualify as fair use).

ii. The Incorporation Test Embraced by P10

From a purely visual perspective, one could define “display” as the mere act of incorporating content into a webpage that is then pulled up by the browser—e.g., the act by SoccerMANIA.com of using an inline link in its webpage to direct the user’s browser to retrieve the Pelé image from SoccerPASSION.com’s server each time he navigates to SoccerMANIA.com. P10 urges the Court to adopt this definition. Under it, SoccerMANIA.com, as the host of its own webpage which incorporates the Pelé photo from SoccerPASSION.com, would be the entity that “displays” that image.

As opposite ends of a spectrum, the server and incorporation tests both are susceptible to extreme or dubious results. Under the server test, someone could create a website entitled “Infringing Content For All!” with thousands of in-line links to images on other websites that serve infringing content. That website, however, would be immune from claims of direct infringement because it does not actually serve the images. On the other hand, under the incorporation test, any website that in-line links to or frames third-party content would risk liability for direct infringement (putting aside the availability of an affirmative defense) even if that website discloses the identity of the actual server of the image. Thus, SoccerMA-NIA.com would expose itself to suit for direct infringement even if the text of its webpage had stated:

To adopt the incorporation test would cause a tremendous chilling effect on the core functionality of the web—its capacity to link, a vital feature of the internet that makes it accessible, creative, and valuable.

iii. Existing Precedents

Only a few courts have addressed the question of whether hyperlinking constitutes “displaying” that infringes a copyright holder’s exclusive right to display his work. Fewer have considered in-line linking or framing.

P10 cites Playboy Enters., Inc. v. Webbworld, Inc., 991 F.Supp. 543 (N.D.Tex.1997). There, Defendant Webbworld, an adult website, received a “news feed” of nude photos from adult internet newsgroups, downloaded them to its computers, and then uploaded them to its own publicly accessible webservers. Id. at 549-50 . The photos included Playboy’s copyrighted images. Webbworld then charged internet users a monthly subscription fee to view the images on its website. Id. at 550 . The district court concluded that Webb-world “displayed” Playboy’s photos because it caused them to be shown on users’ computers and because “[t]he image existed in digital form on Webbworld’s servers.” Id. at 551-52 (emphasis added). Here, it is undisputed that Google does not store or serve any full-size images.

Similarly, in Playboy Enters., Inc. v. Russ Hardenburgh, Inc., 982 F.Supp. 503 (N.D.Ohio 1997), also cited by P10, Defendant Rusty-N-Edie’s, Inc. (“RNE”) operated an electronic bulletin board through its own computers onto which paying subscribers could upload various fries and then receive access to, and the right to download, all the files that other subscribers had uploaded. Id. at 505-506 . When users downloaded files from the bulletin board, those files were transferred to the user’s computer directly from RNE’s computers (not from the original uploader’s computer). The court concluded that RNE had publicly displayed and distributed the files posted on the bulletin board. The court relied, in part, on the fact that after reviewing the files in the upload queue, RNE moved them to its own servers that were available to other subscribers. Id. at 512-13 . Unlike RNE, however, Google does not store or serve any full-size images on Google Image Search.

In Perfect 10, Inc. v. Cybernet Ventures, Inc., 167 F.Supp.2d 1114 (C.D.Cal.2001), Defendant Cybernet Ventures, Inc. (“Cy-bernet”) ran a website called “adult-check.com” which functioned as a gateway to other adult web sites. Paying subscribers would receive access to “the content on any of the related sites within the Adult Check ‘family.’ ” Id. at 1118 . As it does here, Perfect 10 argued that Cybernet directly infringed its copyrighted images. The court denied summary judgment for Cybernet because it lacked sufficient information regarding how Cybernet’s systems interacted with those of its partners, noting “Cybernet may have [had] a direct role in the infringement.” Id. at 1122 . The court did not discuss whether Cybernet had stored or served any of the infringing content. But in a later decision on a motion for preliminary injunction, the same court expressed doubt that liability for direct infringement could be found because Cybernet did not store or serve the infringing content:

Perfect 10, Inc. v. Cybernet Ventures, Inc., 213 F.Supp.2d 1146, 1168-69 (C.D.Cal.2002) (emphasis added).

In its now-withdrawn opinion in Kelly v. Arriba Soft Corp., 280 F.3d 934 (9th Cir.2002) [hereinafter, “Kelly I”], amended by Kelly II, 336 F.3d 811 (9th Cir.2003), the Ninth Circuit discussed liability for direct infringement resulting from in-line linking, without addressing how the technology functioned—i.e., who stored and served the infringing content. Defendant Arriba operated an image search engine much like Google’s—it in-line linked to and framed, but did not store or serve, full-size copies of Kelly’s photographs. Stating that “[n]o cases have addressed the issue of whether inline linking or framing violates a copyright owner’s public display rights,” id. at 945 , the Ninth Circuit nevertheless analogized to Webbworld and Hardenburgh, ignored the fact that the defendants in those two cases actually hosted and served the infringing content, and concluded that Arriba had directly infringed Kelly’s exclusive right to display. Kelly I, 280 F.3d at 945-47 .

Id. at 947 .

The decision in Kelly I was roundly criticized:

4 Nimmer on Copyright [hereinafter “Nimmer”] § 12B.01[A][2] (2005).

Some seventeen months later, perhaps “reflecting] sub silentio that the panel no longer believed in the substance of its much-criticized conclusion,” id., the Ninth Circuit withdrew the portion of the Kelly I opinion dealing with direct infringement on procedural grounds. Kelly II, 336 F.3d at 817 (“[W]e conclude that the district court should not have reached the issue [of whether Arriba’s framing of full-size images constitutes direct infringement] because neither party moved for summary judgment as to the full-size images-”).

Kelly I dealt with an image stored on and served by a third-party website and incorporated into a defendant search engine’s website via in-line linking and framing. Kelly II declined to address whether that conduct constituted direct infringement. Its description of how Arriba functioned is nevertheless useful:

Kelly II, 336 F.3d at 816 . Although such conduct is potentially actionable under secondary liability theories, in terms of direct infringement, Kelly II provides no guidance; the Ninth Circuit just has not settled the question of whether the in-line linking to and framing of content hosted by third-party websites constitutes a “display.”

Certain other decisions, some unpublished, do deal with traditional hyperlink- ing'—ie., rather than incorporating third-party content via in-line linking or framing, websites create hyperlinks that transport the user directly to the linked-to, infringing page. Each of these cases holds that such linking does not implicate any of the exclusive rights under copyright. Ticketmaster Corp. v. Tickets.com, Inc., No. CV 99-7654, 2000 WL 525390 , at *2 (C.D.Cal. Mar. 27, 2000) (unpublished) (“[H]yperlinking does not itself involve a [direct] violation of the Copyright Act (whatever it may do for other claims) since no copying is involved.”); Online Policy Group v. Diebold, Inc., 337 F.Supp.2d 1195 , 1202 n. 12 (N.D.Cal.2004) (“[H]yper-linking per se does not constitute direct copyright infringement because there is no copying, [although] in some instances there may be a tenable claim of contributory infringement or vicarious liability.”); Bernstein v. JC Penney, Inc., No. 98-2958, 1998 WL 906644 , at *1 (C.D.Cal. Sept. 29, 1998) (unpublished) (granting, without discussion, defendant’s motion to dismiss on the ground that hyperlinking cannot constitute direct infringement); Arista Records, Inc. v. MP3Board, Inc., No. 00 CIV. 4660, 2002 WL 1997918 , at *4 (S.D.N.Y. Aug. 29, 2002) (unreported) (linking to content does not implicate distribution right and thus, does not give rise to liability for direct copyright infringement).

These cases, however, are distinguishable because in none of them did defendant actually display anything (or, in the case of Arista Records, distribute anything). In contrast, Google’s in-line linking causes the appearance of copyrighted content on Google’s webpage, even though that content may have been stored on and served by third-party websites.

Hard Rock Cafe Int’l (USA) Inc. v. Morton, No. 97 Civ. 9483, 1999 WL 717995 (S.D.N.Y. Sept. 9, 1999) (unreported), involved trademark, not copyright claims, but is factually analogous. Plaintiff Hard Rock Cafe (“HRCI”) sought to enforce a license agreement with the defendant that precluded the defendant from using the Hard Rock Hotel mark to sell merchandise outside certain permitted means. Granting in part the plaintiffs request, the court ordered the defendant to “cease framing” a website that used the trademark to pro mote the sale of CDs by a company referred to as “Tunes.” The court suggested that in cases involving framed web content, direct trademark infringement may turn on how “smoothly integrated” the two frames are:

Id. at *25 (emphasis added; internal citations omitted). In a footnote, the Court suggested that it might have ruled the other way had defendant’s website (whose webpage created the frame structure), “indicate[d] to the user that he or she ha[d] retrieved a ... web page run by a separate company ....” Id. n. 16 .

Trademark infringement typically concerns issues not applicable to. copyright infringement, such as market confusion and passing off. Thus, it does not necessarily follow from Hard Rock that in assessing whether, for purposes of direct copyright infringement, a work has been “displayed” in a webpage, courts must look to how seamless the transition is between the two frames.

iv. “Display” for Purposes of Full Size Images

The Court concludes that in determining whether Google’s lower frames are a “display” óf infringing material, the most appropriate test is also the most straightforward: the website on which content is stored and by which it is served directly to a user, not the website that in-line links to it, is the website that “displays” the content. Thus, the Court adopts the server test, for several reasons.

First, this test is based on what happens at the technological-level as users browse the web, and thus reflects the reality of how content actually travels over the internet before it is shown on users’ computers. Persons who view the full-size “image in its original context” (ie., the lower frame) after clicking on one of the thumbnails that Google Image Search aggregated, are not viewing images that Google has stored or served. Rather, their computers have engaged in a direct connection with third-party websites, which are themselves responsible for transferring content.

Second, adoption of the server test neither invites copyright infringing activity by a search engine such as Google nor flatly precludes liability for such activity. This test will merely preclude search engines from being held directly liable for in-line linking and/or framing infringing content stored on third-party websites. Copyright owners may still seek, as P10 does, to impose contributory or vicarious liability on websites for the inclusion of such content. Such secondary liability will require analysis of the different set of factors discussed in Section III.B.3 of this Order.

Third, website operators can readily understand the server test and courts can apply it relatively easily. To be sure, the incorporation test, which would have courts look at the URL displayed in the browser’s address bar, also can be applied relatively easily. But that test fails to acknowledge the interconnected nature of the web, both in its physical and logical connections and in its ability to aggregate and present content from multiple sources simultaneously.

Fourth, here the initial direct infringers are the websites that stole PlO’s full-size images and posted them on the internet for all the world to see. P10 would not have filed suit but for their actions.

Finally, the server test maintains, however uneasily, the delicate balance for which copyright law strives—i.e., between encouraging the creation of creative works and encouraging the dissemination of information. Merely to index the web so that users can more readily find the information they seek should not constitute direct infringement, but to host and serve infringing content may directly violate the rights of copyright holders.

Applying the server test, the Court concludes that for the purposes of direct copyright infringement, Google’s use of frames and in-line links does not constitute a “display” of the full-size images stored on and served by infringing third-party websites. Thus, PlO’s claim of direct infringement with respect to these actions will likely fail.

c. “Display” for Purposes of Thumbnails

Applying the server test to Google’s use of thumbnails the Court finds that Google does “display” thumbnails of PlO’s copyrighted images. Google acknowledges that it creates and stores those thumbnails on its own servers—and that upon receiving search queries, it responds by displaying a grid of those thumbnails.

d. What Constitutes a “Public Distribution”?

The foregoing considerations also inform whether Google directly infringes PlO’s distribution right. With respect to PlO’s full-size images, Google does not. A distribution of a copyrighted work requires an “actual dissemination” of copies. See In re Napster, Inc. Copyright Litigation, 377 F.Supp.2d 796, 802-804 (N.D.Cal.2005); accord Nimmer § 8.11[A]. In the internet context, an actual dissemination means the transfer of a file from one computer to another. Although Google frames and in-line links to third-party infringing websites, it is those websites, not Google, that transfer the full-size images to users’ computers. Because Google is not involved in the transfer, Google has not actually disseminated—and hence, and has not distributed—the infringing content. See In re Napster, 377 F.Supp.2d at 802-804 (N.D.Cal.2005) (finding that Napster had not “distributed” songs in light of the fact that the “infringing works never resided on the Napster system,” and therefore, Napster could not have transferred copyrighted content to its users).

Accordingly, the Court concludes that by merely framing and in-line linking to third- party websites, Google has not “distributed” infringing copies of PlO’s copyrighted full-size photographs.

2. Fair Use

Having found that the thumbnails directly infringe PlO’s copyrights, the Court turns to Google’s affirmative defense of fair use. Google argues that its creation and display of thumbnails is fair under 17 U.S.C. § 107 . “From the infancy of copyright protection, some opportunity for fair use of copyrighted materials has been thought necessary to fulfill copyright’s very purpose, ‘[t]o promote the Progress of Science and useful Arts.... ’ U.S. Const., Art. I, § 8, cl. 8.” Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 575 , 114 S.Ct. 1164 , 127 L.Ed.2d 500 (1994). This notion was codified in 17 U.S.C. § 107 :

Although often discussed within the context of the first factor, “the public interest is also a factor that continually informs the fair use analysis.” Nimmer § 13.05[B][4]; see also Sony Computer Entm’t America, Inc. v. Bleem, LLC, 214 F.3d 1022, 1027 (9th Cir.2000); Sega Enters. Ltd. v. Accolade, Inc., 977 F.2d 1510, 1523 (9th Cir.1992) (“[W]e are free to consider the public benefit resulting from a particular use.... ”). Courts are to consider “these factors in light of the objectives of copyright law, rather than view them as definitive or determinative tests.” Kelly II, 336 F.3d at 818 .

a. Purpose and Character of Use

“[T]he preamble to Section 107 ... enumerate^] certain purposes that are most appropriate for a finding of fair use: ‘criticism, comment, news reporting, teaching!,] scholarship or research.’ ” Nimmer § 13.05[A][l][a]. “The central purpose of [the first fair use factor] is to see ... whether the new work merely supersedes the objects of the original creation, or instead adds something new, with a further purpose or different character, altering the first with new expression, meaning, or message; it asks, in other words, whether and to what extent the new work is ‘transformative.’ ” Campbell, 510 U.S. at 579 , 114 S.Ct. 1164 . Although the Supreme Court has stated that “every commercial use of copyrighted material is presumptively an unfair exploitation of the monopoly privilege that belongs to the owner of the copyright,” Sony, 464 U.S. at 451 , 104 S.Ct. 774 , this pronouncement does not preclude a finding that a defendant’s commercial use may nevertheless be fair. Kelly II, 336 F.3d at 818 (citing Campbell, 510 U.S. at 579, 114 S.Ct. 1164 ). Furthermore, “[t]he more transformative the new work, the less important the other factors, including commercialism, become.” Id.

i. Commercial Versus Noncommercial Use

In assessing whether a use is commercial, the focus here is not on the individuals who use Google Image Search to locate PlO’s adult images. Nor is it on whether their subsequent use of the images is noncommercial {e.g., titillation) or commercial {e.g., to print and sell). Rather, it is Google’s use that the Court is to consider. That use, P10 contends, is commercial in nature. The Court agrees.

Courts have defined “commercial uses” extremely broadly. See Nimmer § 13.05[A][l][c] (providing examples). Google unquestionably derives significant commercial benefit from Google Image Search in the form of increased user traffic'—and, in turn, increased advertising revenue. The more people who view its pages and rely on its search capabilities, the more influence Google wields in the search engine market and (more broadly) in the web portal market. In turn, Google can attract more advertisers to its Ad-Sense and AdWords programs.

That Google’s use of thumbnails is commercial, however, does not necessarily weigh heavily in favor of P10. In Kelly II, supra, the Ninth Circuit acknowledged that Arriba’s use was commercial, yet concluded that that fact “weighted] only slightly against a finding of fair use” because “Arriba was neither using Kelly’s images to directly promote its web site nor trying to profit by selling Kelly’s images.” Kelly II, 336 F.3d at 818 . The court found that the creation and use of thumbnails to display Arriba’s image search results were “more incidental and less exploitative in nature than more traditional types of commercial use.” Id.

Google Image Search automatically scours the internet via its web crawler software to find and catalog images. For each image, Google records information about it and creates a thumbnail copy. The thumbnail is then stored in Google’s cache for later display when users perform an image search. When an image search is performed, Google displays a grid of thumbnails that Google has algorithmically determined are responsive to the search string based on the text of the originating webpage surrounding the image. When a user clicks on one of the thumbnails, he is taken to the two-frame page discussed above. In these respects, Google functions like Arriba’s search engine.

But unlike Arriba, Google offers and derives commercial benefit from its Ad-Sense program. AdSense allows' third party websites “to carry Google-sponsored advertising and share revenue that flows from the advertising displays and click-throughs.” Def.’s MacGillivray Decl. ¶ 10. If third-party websites that contain infringing copies of P10 photographs are also AdSense partners, Google will serve advertisements on those sites and split the revenue generated from users who click on the Google-served advertisements. Google counters that its AdSense Program Policies prohibit a website from registering as an AdSense partner if the site’s webpages contain images that appear in Google Image Search results: “In order to avoid associations with copyright claims, website publishers may not display Google ads on web pages with ... Image Results.” Id. ¶ 11. However, Google has not presented any information regarding the extent to which this purported policy is enforced. Nor has it provided examples of AdSense partners who were terminated because of violations of this policy. In contrast, P10 has submitted numerous screenshots of third-party websites that serve infringing content and also appear to be receiving and displaying AdSense ads from Google. Pl.’s Zada Decl., Exs. 28-29.

AdSense unquestionably makes Google’s use of thumbnails on its image search far more commercial than Arriba’s use in Kelly II. Google’s thumbnails lead users to sites that directly benefit Google’s bottom line. Id., Ex. 6 (Second Quarter Fiscal 2005 Report) at 98 (“Revenues generated on Google’s partner sites, through Ad-Sense programs, contributed $630 million, or 46% of total revenues.... ” ). Google has a strong incentive to link to as many third-party websites as possible—including those that host AdSense advertisements,

ii. Transformative Versus Consumptive Use

That a use is commercial does not preclude a defendant from tipping the balance back to a finding of fair use by showing that its use is “transformative,” as opposed to “consumptive.” A consumptive use is one in which defendant’s “use of the images merely supersede^] the object of the originals ... instead [of] add[ing] a further purpose or different character.” Kelly II, 336 F.3d at 818 . Whether a use is transformative depends in part on whether it serves the public interest. See Religious Tech. Ctr. v. Netcom On-Line Communication Servs., 907 F.Supp. 1361, 1379 (N.D.Cal.1995) (“Netcom’s use, though commercial, also benefits the public in allowing for the functioning of the Internet and the dissemination of other creative works, a goal of the Copyright Act.”); Sega Enters., 977 F.2d at 1522-23 (intermediate copying to reverse engineer software is fair use .despite commercial nature of activity in light of public benefit).

P10 argues that Google’s use of thumbnails is consumptive rather than transfor-mative since Google “provides the exact same images through the exact same medium ... as does Perfect 10.” Whether thumbnails are identical copies of their full-size counterparts is debatable. A thumbnail contains significantly less pixel data (and hence, less image detail) than does the full-size image. The more complex or nuanced the original full-size image, the less exact is the replicated viewing experience—i.e., at some point viewers can no longer discern many of the fine details that were once visible in the full-size image. On the other hand, thumbnails are not “cropped” in any way, and if few or no important details have been lost, they do convey the full expression—they achieve pretty much the same effect—as the original full-size images. Merely because Google’s thumbnails are not cropped does not necessarily make them exact copies of PlO’s images, but the record currently before the Court does suggest that the thumbnails here closely approximate a key function of PlO’s full-size originals, at least to the extent that viewers of PlO’s photos of nude women pay little attention to fine details.

Google’s use of thumbnails does not supersede PlO’s use of full-size images. In the final analysis, PlO’s use is to provide “entertainment,” both in magazines and on the internet. For some viewers, PlO’s use of the photos creates or allows for an aesthetic experience. Google, in contrast, does not profit from providing adult content, but from locating, managing, and making information generally more accessible, and therefore more attractive to advertisers. Google is focused almost exclusively on the web and is involved in a wide variety of internet-related projects (e.g., web search, desktop search, newsgroup search, map and directory services, academic research, and language translation services). In this respect, Google’s wide-ranging use of thumbnails is highly trans-formative: their creation and display is designed to, and does, display visual search results quickly and efficiently to users of Google Image Search.

The Ninth Circuit’s opinion in Kelly II is particularly instructive on this point:

Kelly II, 336 F.3d at 818-820 .

It is by now a truism that search engines such as Google Image Search pro vide great value to the public. Indeed, given the exponentially increasing amounts of data on the web, search engines have become essential sources of vital information for individuals, governments, nonprofits, and businesses who seek to locate information. As such, Google’s use of thumbnails to simplify and expedite access to information is transformative of PlO’s use of reduced-size images to entertain. But that does not end the analysis, because Google’s use is simultaneously consumptive as well. In early 2005, after it filed suit against Google, P10 entered into a licensing agreement with Fonestarz Media Limited for the sale and distribution of P10 reduced-size images for download to and use on cell phones. Google’s use of thumbnails does supersede this use of PlO’s images, because mobile users can download and save the thumbnails displayed by Google Image Search onto their phones. Google’s thumbnail images are essentially the same size and of the same quality as the reduced-size images that P10 licenses to Fonestarz. Hence, to the extent that users may choose to download free images to their phone rather than purchase PlO’s reduced-size images, Google’s use supersedes PlO’s.

In Kelly II, the Ninth Circuit found the first fair use factor to weigh in favor of Arriba. Kelly II, 336 F.3d at 820 (“[T]his first factor weighs in favor of Arriba due to the public benefit of the search engine and the minimal loss of integrity to Kelly’s images.”). Here, because Google’s use of thumbnails is more commercial than Arri-ba’s and because it is “consumptive” with respect to PlO’s reduced-size images, the Court concludes that this factor weighs slightly in favor of P10.

b. Nature of Copyrighted Work

“ ‘Works that are creative in nature are closer to the core of intended copyright protection than are more fact-based works.’ ” Kelly II, 336 F.3d at 820 (citing A & M Records, 239 F.3d at 1016 ). “Photographs that are meant to be viewed by the public for informative and aesthetic purposes ... are generally creative in nature.” Id.; accord Elvis Presley, 349 F.3d at 629 (“[P]hotographs taken for aesthetic purposes, are creative in nature and thus fit squarely within the core of copyright protection.”).

“The fact that a work is published or unpublished also is a critical element of its nature” given a copyright holder’s right of first publication. Kelly II, 336 F.3d at 820 (citing Harper & Row Publishers, Inc. v. Nation Enters., 471 U.S. 539, 564 , 105 S.Ct. 2218 , 85 L.Ed.2d 588 (1985)). “Published works are more likely to qualify as fair use because the first appearance of the artist’s expression has already occurred.” Id.

In Kelly II, however, the Ninth Circuit found that Kelly’s photographs, although creative, had “appeared on the internet before Arriba used them in its search image.” Kelly II, 336 F.3d at 820 . Partly for that reason, the Ninth Circuit concluded that although the second statutory fair use factor weighed in favor of Kelly, its weight was slight. Id. Similarly, here, although PlO’s images are “creative,” they, too, have previously been published, both in print and on the web. Thus, as in Kelly II, the Court concludes that this factor weighs only slightly in favor of P10.

c. Amount and Substantiality of the Portion Used

“While wholesale copying does not preclude fair use per se, copying an entire work militates against a finding of fair use.” Kelly II, 836 F.3d at 820 (citing Worldwide Church of God v. Philadelphia Church of God, Inc., 221 F.3d 1110, 1118 (9th Cir.2000)). “However, the extent of permissible copying varies with the purpose and character of the use.” Id. (citing Campbell, 510 U.S. at 586-87 , 114 S.Ct. 1164 ). “If the secondary user only copies as much as is necessary for his or her intended use, then this factor will not weigh against him or her.” Id. at 820-821.

Id. at 821.

The Court finds that Google’s use of the infringing copies of PlO’s images also is no greater than necessary to achieve the objective of providing effective image search capabilities. In doing so, the Court rejects PlO’s contention that Google could have provided such assistance through the use of text, claiming that PlO’s images are more readily describable in words than Kelly’s images. First, contrary to PlO’s contention, photographs of nude women can, like photographs of the American West, vary greatly. Second, both kinds of pictures can be described verbally, yet no matter how susceptible any image is to textual description words cannot adequately substitute for thumbnails in quickly and accurately conveying the content of indexed full-size images.

Thus, as in Kelly II, the Court finds that this factor favors neither party.

d. Effect of the Use upon the Potential Market for and Value of the Copyrighted Work

“This last [fair use] factor requires courts to consider not only the extent of market harm caused by the particular actions of the alleged infringer, but also whether unrestricted and widespread conduct of the sort engaged in by the defendant ... would result in a substantially adverse impact on the potential market for the original.” Kelly II, 336 F.3d at 821 (internal quotations omitted). “A transfor-mative work is less likely to have an adverse impact on the market of the original than a work that merely supersedes the copyrighted work.” Id.

P10 targets its copyrighted small- and full-size images at several markets: the print magazine market, the online adult website subscription market, and the cell phone image download market. Google’s use of thumbnails is not likely to affect the market for full-size images (whether in print or online). As stated in Kelly II,

Id.

On the other hand, Google’s use of thumbnails likely does harm the potential market for the downloading of PlO’s reduced-size images onto cell phones. Google argues that because “P10 admits [that] this market is growing,” its “delivery of thumbnail search results” must not be having a negative impact. Apart from being more relevant to the quantification of damages, this weak argument overlooks the fact that the cell phone image-download market may have grown even faster but for the fact that mobile users of Google Image Search can download the Google thumbnails at no cost. Commonsense dictates that such users will be less likely to purchase the downloadable P10 content licensed to Fonestarz.

e. Conclusion Regarding Fair Use and Direct Infringement

The first, second, and fourth fair use factors weigh slightly in favor of P10. The third weighs in neither party’s favor. Accordingly, the Court concludes that Google’s creation of thumbnails of PlO’s copyrighted full-size images, and the subsequent display of those thumbnails as Google Image Search results, likely do not fall within the fair use exception. The Court reaches this conclusion despite the enormous public benefit that search engines such as Google provide. Although the Court is reluctant to issue a ruling that might impede the advance of internet technology, and although it is appropriate for courts to consider the immense value to the public of such technologies, existing judicial precedents do not allow such considerations to trump a reasoned analysis of the four fair use factors.

To summarize, then: (1) at this stage PIO has not established that it is likely to prove that Google’s framing of and in-line linking to infringing (full-size) copies of PlO’s images constitutes a public display or distribution rendering Google liable for direct infringement; but (2) PIO has established a likelihood of proving that Google’s creation and public display of thumbnails does directly infringe PlO’s copyrights.

That PIO has established a likelihood of success on one facet of its direct infringement claims does not necessarily mean that it is entitled to all of its proposed injunctive relief. Because certain aspects of that relief relate to conduct of Google that must be addressed under principles of secondary liability, the Court now turns to PlO’s remaining claims.

3. Secondary Copyright Liability— Contributory and Vicarious Infringement

P10 contends that Google is likely to be held secondarily liable under the doctrines of contributory and vicarious infringement. “One infringes contributorily by intentionally inducing or encouraging direct infringement ... and infringes vicariously by profiting from direct infringement while declining to exercise a right to stop or limit it....” Metro-Goldwyn-Mayer Studios Inc. v. Grokster, Ltd., — U.S. -, 125 S.Ct. 2764, 2776 , 162 L.Ed.2d 781 (2005) [hereinafter “Grok ster”] (citations omitted).

P10 argues that parties other than Google directly infringe its copyrights in two ways: (1) third-party websites directly infringe by reproducing, displaying, and distributing unauthorized copies of PlO’s copyrighted photographs and (2) users of Google directly infringe by downloading such images, thereby making infringing reproductions. Google does not contest that numerous third-party websites directly infringe by serving PlO’s copyrighted images. However, Google does argue that PIO has presented no evidence indicating that individual users of Google engage in direct infringement upon finding copyrighted P10 photos on the web. Google contends that “[t]here are countless ways Google searchers can ‘use’ Google’s search results, including fair uses, and Perfect 10’s evidence is missing on this point.” On this point, the Court agrees with Google. PIO has not submitted evidence showing that individual users of Google themselves infringe PlO’s copyrights. PIO has demonstrated only that users of Google search are capable of directly infringing by downloading the underlying webpage or image. It is not unlikely that many users do just that, but on this preliminary injunction motion there is no evidence in the record proving so. In contrast, in the Napster and Grokster cases, there was overwhelming evidence that on a massive scale file-sharers were using those defendants’ software (essentially, peer-to-peer music search engines) to download, and thereby directly infringe, copyrighted works. See, e.g., A & M Records, Inc. v. Napster, Inc., 239 F.3d 1004, 1013-14 (9th Cir.2001) [hereinafter “Napster II”]; Grokster, 125 S.Ct. at 2772 . Furthermore, in those cases the file-sharers actually had to download songs in order to enjoy the music, thereby making infringing reproductions. In contrast, to view PlO’s photos, users of Google’s search engine need only visit the third-party website that hosts, and serves the infringing adult content.

P10’s arguments that Google is secondarily liable therefore must be assessed in light of the only direct infringement (other than as to thumbnails) for which there is evidence: that of third-party websites that reproduce and display unauthorized copies of PlO’s photographs. As to these websites’ actions, P10 argues that Google is aware of, materially contributes to, profits from, and declines to supervise such direct infringement by (1) providing infringing websites an “audience” (by helping users locate them) and (2) providing a revenue stream to infringing websites via AdSense.

a. Contributory Infringement

To substantiate its claim of contributory infringement, P10 must show (1) that Google had knowledge of the infringing activity and (2) that Google induced, caused, or materially contributed to that activity. Ellison v. Robertson, 357 F.3d 1072, 1076 (9th Cir.2004).

Google argues that it cannot be held contributorily liable under the Supreme Court’s holding in Sony Corp. of America v. Universal City Studios, Inc., 464 U.S. 417 , 104 S.Ct. 774 , 78 L.Ed.2d 574 (1984). That seminal 1984 Supreme Court decision “barred secondary liability based on presuming or imputing intent to cause infringement solely from the design or distribution of a product capable of substantial lawful use, which the distributor knows is in fact used for infringement.” Grokster, 125 S.Ct. at 2778 (paraphrasing Sony’s holding). Under Sony , Google cannot be deemed to have constructive knowledge of infringing activity since its search engine clearly is capable of commercially significant noninfringing uses. However, “where evidence goes beyond a product’s characteristics or the knowledge that it may be put to infringing uses, and shows statements or actions directed to promoting infringement, Sony’s ... rule will not preclude liability.” Id. at 2779. As the Ninth Circuit presciently stated in Napster II: “[A]ctual, specific knowledge of direct infringement renders Sony’s holding of limited assistance” to a defendant seeking to avoid contributory liability. Napster II, 239 F.3d at 1020 .

i. Knowledge

Whether an alleged secondary infringer has “knowledge” of an infringing activity includes both actual knowledge and constructive knowledge: “Contributory liability requires that the secondary infringer ‘know or have reason to know' of direct infringement.” Id.

P10 contends that Google has actual knowledge of specific acts of infringement based upon (1) numerous notices of infringement that P10 has sent Google, (2) the fact that certain infringing images contain “Perfect 10 copyright notices, or labels such as ‘P10 Fall 1999,’ ” and (3) the fact that Google “monitors the content of allegedly Infringing Sites.”

The Court rejects PlO’s second and third contentions. Google does not necessarily know that any given image on the internet is infringing someone’s copyright merely because the image contains a copyright notice. Google would need more information in order to know whether the operator of that third-party website created the work, licensed its use or instead was illegally displaying an infringing copy of it. Nor does Google obtain actual knowledge of infringement as a result of its alleged monitoring practices. Google claims that it no longer follows the Ad-Sense policies to which P10 points. The current AdSense Program Policies web-page no longer contains language reserving to Google the right to “monitor” its AdSense partners. In any event, merely because Google may have reserved the right to monitor its AdSense partners does not mean that it could thereby discern whether the images served by those websites were subject to copyright. Only upon receiving proper notice of alleged infringement can Google determine whether a given AdSense partner has violated the terms of Google’s AdSense Program Policies.

Thus, the question here is whether P10 provided Google with adequate actual knowledge of specific infringing activities. Dr. Zada began to send “notices of infringement” in May 2001 and continued to do so through 2005. PL’s Zada Decl. ¶¶ 76-89. Google acknowledges that it received PlO’s notices. Def.’s MacGillivray Decl. ¶ 19. It argues, however, that those notices frequently did not describe in sufficient detail the specific location (URL) of an infringing image and frequently did not identify the underlying copyrighted work. Id. ¶20. For example, some notices included invalid, truncated URLs with an ellipsis between the domain name of the website and the file name of the particular webpage on which that image appeared. See, e.g., id., Ex. E at 131. Other notices listed entire websites as infringing, or entire directories within a website. Google claims that despite these shortcomings, it promptly processed all of the notices it received, suppressing links to specific web-pages that it could confirm displayed infringing P10 photos. Id. ¶¶ 21-25.

In the next section the Court concludes that Google does not materially contribute to the direct infringement of PlO’s photos by third-party websites. For that reason, without actually deciding the question, the Court will assume that Google has actual knowledge of infringement and will proceed to analyze the second element of contributory liability—i.e., actual furthering of the activity.

ii. Material Contribution

To materially contribute to directly infringing activity, the defendant must “engage[ ] in ‘personal conduct that encourages or assists the infringement.’ ” Napster II, 239 F.3d at 1019 . P10 contends that Google does so by (1) giving infringing websites an audience (ie., allowing users to locate infringing sites) and (2) providing infringing websites with a revenue stream via AdSense or increasing their existing revenue stream by increasing user traffic.

For legal support that what Google does “materially contributes,” P10 relies heavily on the Ninth Circuit’s decision in Napster II. In light of the plentiful factual distinctions between Napster II and this case (as detailed in the following chart), PlO’s reliance on that decision is misplaced.

[[Image here]]

[[Image here]]

Napster II, 239 F.3d at 1011-12 ; Napster I, 114 F.Supp.2d at 905-908, 920 .

Google resembles Napster only in facilitating searches (i.e., helping users find information)—and even then there are significant differences. Whereas Napster dedicated itself to helping users locate audio files found on the otherwise inaccessible hard drives of individual users, Google helps users locate all types of information (text, images, video, newsgroup discussion threads, blogs, academic papers, price information, maps, driving directions) found on the entire, publicly accessible web.

In short, Google does not materially contribute to direct infringement in the ways or to the extent that Napster did. Nevertheless, P10 argues, Google contributes differently: it provides an “audience” and brand recognition for infringing third-party websites and it advertises for the sites. P10 overstates Google’s actual conduct and confuses search technology with active encouragement and promotion of infringing activity. P10 likens this case to Gershwin Publ’g. Corp. v. Columbia Artists Mgmt., Inc., 443 F.2d 1159 (2d Cir.1971). There, the Second Circuit found material contribution in defendant’s “pervasive participation in the formation and direction of [an] association” which put on concerts at which copyrighted musical compositions were performed. Id. at 1163 . The defendant organized the concerts, helped select and book artists, prepared budgets and artist contracts, created publicity kits, held one-week membership campaigns and compiled a report of the proceeds of the concerts. Id. at 1160-61 . Nothing Google does is comparable.

PlO’s citation to Columbia Pictures Indus., Inc. v. Redd Horne, Inc., 749 F.2d 154 (3d Cir.1984) is similarly inapposite. In Columbia Pictures, the defendant, unlike Google, “conducted all of the advertising and promotional work for [the alleged direct infringer, and also] provided financial, accounting, and administrative services .... All of these services, and the advertising services in particular, contributed and, indeed, were essential to the copyright infringement.” Id. at 161 . Here, in contrast, Google has not actively encouraged users to visit infringing third-party websites, and it has not induced or encouraged such websites to serve infringing content in the first place. Moreover, it would be a “gross generalization that cannot withstand scrutiny” to argue that “supplying the ‘means’ to accomplish an infringing activity and [even] encouraging that activity through advertisement are sufficient to establish liability for copyright infringement.” Sony, 464 U.S. at 436 , 104 S.Ct. 774 (emphasis added).

P10 cites Grokster in support of its claim that Google materially contributes by “providing] a revenue stream to infringing websites, including placing [AdSense] advertisements next to infringing P10 images on these third party websites.” In the language that P10 cites, the Supreme Court noted that StreamCast and Grokster made money

Grokster, 125 S.Ct. at 2781-82 . This language, however, pertained to the Supreme Court’s “inducement theory” of contributory infringement, not to the analysis of “material contribution.” Although the two inquiries overlap to some extent, id. at 2782 n. 13, Grokster does not support PlO’s argument that AdSense materially contributes to direct infringement occurring on third-party websites. Although AdSense may provide some level of additional revenue to these websites, PIO has not presented any evidence establishing what that revenue is, much less that it is material, (either in its own right or relative to those websites’ total income). There is no evidence that these sites rely on Google Ad-Sense for their continued existence or that they were created with the purpose of profiting from the display of AdSense advertisements.

In short, PIO has failed to meet its burden of establishing that it is likely that Google’s AdSense program will be found to materially contribute to the direct infringement taking place on infringing third-party websites. Such websites existed long before Google Image Search was developed and would continue to exist were Google Image Search shut down. Accordingly, it is unlikely that Google will be found eon-tributorily liable.

b. Vicarious Infringement

To prove vicarious infringement, P10 must show (1) that Google enjoys a direct financial benefit from the infringing activity of third-party websites that host and serve infringing copies of P10 photographs and (2) that Google has declined to exercise the right and ability to supervise or control the infringing activi ty. Ellison v. Robertson, 357 F.3d 1072, 1076 (9th Cir.2004).

Neither party has submitted any evidence, such as economic data or expert reports, showing the extent (if any) to which Google benefits financially from copyright infringement by third-party websites. It is likely that at least some users are drawn to Google Image Search because they know that copies of PlO’s photos can be viewed for free, and it is indisputable that Google does stand to benefit the more users visit and use Google Image Search. In principle, any increase in Google’s web traffic leads to increased advertising revenue, brand awareness, and market clout for Google. Does that constitute a “direct” financial benefit?

In Napster II, the Ninth Circuit found that Napster had a direct financial interest in the infringing activity, and, citing to Fonovisa, Inc. v. Cherry Auction, Inc., 76 F.3d 259 (9th Cir.1996), the Court of Appeals stated,

Napster II, 239 F.3d at 1023 . This broad definition of “direct financial benefit” would encompass even a “future hope to ‘monetize.’ ” Nimmer § 12.04[A][1] (commenting on Napster II). Under this standard, Google clearly benefits financially from third parties’ displays of PlO’s photos. Google certainly derives a direct financial benefit if users visit AdSense partners’ websites that contain such infringing photos. If Google serves advertisements that are displayed on such sites, it will share in the ad revenue. Hence, its financial benefit is direct.

As to the second prong of vicarious liability, P10 again invokes Napster II in support of its contention that Google has the right and ability to control the infringing activity taking place on the web. In Napster II, the Ninth Circuit stated that “the ability to block infringers’ access to a particular environment for any reason whatsoever is evidence of the right and ability to supervise.” Napster II, 239 F.3d at 1023 (emphasis added). Napster did control the “particular environment” in which its file-sharing service operated; its architecture was based on a proprietary, closed-universe system, not an open, web-based service. If Napster removed a link to infringing content, the content was no longer available on Napster’s entire file-sharing network. That is not the case here. Google does not exercise control over the environment in which it oper ates—i.e., the web. Google’s ability to remove a link from its search index does not render the linked-to site inaccessible. The site remains accessible both directly and indirectly (i.e., via other search engines, as well as via the mesh of websites that link to it). If the phrase “right and ability to control” means having substantial input into or authority over the decision to serve or continue serving infringing content, Google lacks such right or ability.

Moreover, Google’s software lacks the ability to analyze every image on the internet, compare each image to all the other copyrighted images that exist in the world (or even to that much smaller subset of images that have been submitted to Google by copyright owners such as PIO), and determine whether a certain image on the web infringes someone’s copyright. Def.’s Levine Decl. ¶ 22. In addition, Google’s right and ability to remove infringing websites from its index would make it more difficult for such websites to be found on the web, but those sites would continue to exist anyway. Google cannot shut down infringing websites or prevent them from continuing to provide infringing content to the world.

P10 points out that Google’s AdSense policies reserve the right to monitor and terminate partnerships with entities that violate others’ copyright. This, P10 contends, is evidence of Google’s right and ability to control AdSense partners’ infringing conduct. In Hendrickson v. eBay, Inc., 165 F.Supp.2d 1082 (C.D.Cal.2001), the court considered the issue in the context of the DMCA § 512(c) safe harbor (ie., for information residing on systems or networks at the direction of users), held that “the ‘right and ability to control’ ... infringing activity ... cannot simply mean the ability of a service provider to remove or block access to materials posted on its website or stored in its system.” Id. at 1093-94. The court rejected plaintiffs argument that eBay’s “ability to control” arose from its ability to remove infringing auction listings. Id. Accord, Perfect 10 v. CCBill, LLC, 340 F.Supp.2d 1077, 1098 (C.D.Cal.2004) (right and ability to “disconnect[ ] access ... is not sufficient ... to demonstrate a ‘right and ability to control’ the infringing activity”).

The “right and ability to control” infringing activity of others requires more than what is embodied in Google’s Ad-Sense revenue-sharing agreement; there must be some form of control over or authority to stop or limit the infringing conduct itself. P10 has not established a likelihood of proving the second prong necessary for vicarious liability.

4. Conclusion Regarding Likelihood of Success

P10 is likely to succeed in proving that Google directly infringes by creating and displaying thumbnail copies of its photographs. P10 is unlikely to succeed in proving that Google can be held secondarily liable.

C. Irreparable Harm

In copyright cases, irreparable harm is presumed once a sufficient likelihood of success is raised. Elvis Presley, 349 F.3d at 627 . Google argues that P10 first sent notices of infringement in May 2001 and then “waited three and a half years before filing this lawsuit [and] another nine months to seek a preliminary injunction.” Google contends that this constitutes “unreasonable delay” and rebuts any presumption of immediate or irreparable harm. Although P10 did begin sending notices of infringement as early p as May 2001, those notices concerned solely Google Web Search. See Pl.’s Zada Decl. ¶ 76. P10 was not aware until May 2004 that Google displayed thumbnails of PlO’s copyrighted images on Google Image Search. Id. ¶ 79. Shortly thereafter, P10 began sending notices of infringement. Although P10 did wait six months to file suit and another nine months to seek a preliminary injunction, P10 did so justifiably; it was engaged in settlement discussions with Google and was evaluating whether Google would remove the infringing thumbnail images from its index. P10 has satisfied the “irreparable harm” element.

D. Public Interest

Google argues that the “value of facilitating and improving access to information on the Internet ... counsels against an injunction here.” This point has some merit. However, the public interest is also served when the rights of copyright holders are protected against acts likely constituting infringement. Furthermore, in this case a preliminary injunction can be carefully tailored to balance the competing interests described in the first paragraph of this Order: those of intellectual property rights on the one hand and those promoting access to information on the other. The Court ORDERS P10 and Google to propose jointly the language of such an injunction, and to lodge their proposal by not later than March 8, 2006.

IV. CONCLUSION

For the reasons discussed above, the Court GRANTS IN PART and DENIES IN PART PlO’s motion for a preliminary injunction against Google.

IT IS SO ORDERED.

[[Image here]]

. A9.com is Amazon.com’s search website. . In this Order, Declarants proffered by Plaintiff will be referred to as "Pl.’s ___ Decl.” Defendants’ declarants will be referred to as "Def.’s - Decl.” Norman Zada (formerly Zadeh) is the founder and CEO of P10. Pl.’s Zada Decl. ¶ 2. Zada received a Ph.D. in Operations Research from the University of California at Berkeley in 1972. Id. ¶ 3. From 1972 to 1973, Zada performed computer science research for IBM. Id. Zada has also taught applied mathematics as a visiting professor at Stanford University, UCLA, U.C. Irvine, and Columbia University. Id. ¶ 4. . Subscribers choose a unique user-name/password combination which allows them to log into the "members’ area.” .A "thumbnail” is a lower-resolution (and hence, smaller) version of a full-size image. Thumbnails enable users to quickly process and locate visual information. For example, users of Google Image Search are presented with a set of thumbnails that are potentially responsive to their search queries. Because thumbnails are smaller in size, more of them can be displayed at the same time on a single page or screen. Users can quickly scan the entire set of thumbnails to locate the particular full-size image for which they were looking. P10 repeatedly objects that the term "thumbnail” is a misnomer, even going so far as to point out that the thumbnails displayed by Google can be up to eight times the size of a person’s actual thumbnail. PL's Zada Reply Decl. ¶ 54. "Thumbnail,” it argues, conveys the false impression that smaller, lower-resolution images are not useful in and of themselves—or that they are less useful than their full-size counterparts. The term "thumbnail,” however, has become the standard way of referring to the smaller, lower-resolution images central to this suit. In any event, the Court recognizes that thumbnails have been used for purposes independent of their primary function, as is discussed later. See, e.g., Kelly v. Arriba Soft Corp., 336 F.3d 811, 815 (9th Cir.2003). . Google Image Search does not have the ability to accept an image as a search query and return similar images. Only text-based search queries can be input. Google Image Search returns those images on the internet whose surrounding text was deemed responsive to the user's textual search string. . Since URLs may often be extremely long, Google displays the domain name of the third-party website and the file name of the image, but the middle portion of the URL frequently contains an ellipsis indicating that the full URL has been truncated. . To illustrate how AdSense works, an individual who maintains a website dedicated to soccer—' 'SoccerMANIA.com,' ’ say—might post his personal commentary about recent games, along with player profiles and a short history of soccer. What he is not likely to do—perhaps because it is time-consuming or outside his area of expertise, or simply because he does not choose to—is find advertisers who are willing to pay to place advertisements on his site. This is where Google AdSense comes in. After registering to become an AdSense partner, the soccer aficionado can demarcate an area on his website that acts as a "placeholder” for an advertisement. Google will then scan the text of his website and populate or fill the placeholder with advertisements it deems relevant to the content on that site. Google’s AdSense software will notice that the word "soccer” and other soccer-related terms appear frequently on the site, and thus will show advertisements directed at people interested in soc cer—e.g., sites that sell tickets to World Cup games. . That factor is "the effect of the use upon the potential market for or value of the copyright-edwork.” 17 U.S.C. § 107 . . P10 also argues that Google's web (text-based) search function directly infringes PlO’s intellectual property by linking to websites that display the username/password combinations that subscribers to perfectlO.com have obtained in order to gain access to the "members’ only” area of the site. The Court rejects this argument. PIO has not demonstrated that it has any copyright interest in the two strings of characters that other individuals select when registering as members on perfect], 0.com. Furthermore, as explained more fully below, Google does not risk liability for direct infringement merely by linking to content hosted on and served by third-party websites. . That website, however, might still be held liable for secondary infringement. See below. . P10 also argues that Google "distributes” the thumbnails. Although Google does transfer thumbnails to users' computers though local browser caching, this automatic "distribution” likely constitutes fair use. See note 17 infra. In any event, because the Court concludes that Google’s creation and display of thumbnails directly infringes PlO's copyrights, the question of whether Google also distributes those thumbnails is moot. . Of course, this figure represents the gross revenue earned through AdSense. It is not discounted by the amount paid back to Ad-Sense partners; nor does it break down the much smaller amount attributable to websites that contain infringing content. . For example, a typical full-size image might be 1024 pixels wide by 768 pixels high, for a total of 786, 432 pixels worth of data. A typical thumbnail might be 150 pixels wide by 112 pixels high, for a total of only 16,800 pixels. This represents an information loss of 97.9% between the full-size image and the thumbnail. . This inquiry is closely related to the fourth fair use factor, i.e., the impact on plaintiff's potential market. . Google argues that PlO's works are not creative because P10 “emphasizes the objects of the photographs (nude women) and [P10] assumes that persons seeking Perfect 10’s photos are searching for the models and for sexual gratification.” Google contends that this "implies a factual nature of the photographs.” The Court rejects this argument. The P10 photographs consistently reflect professional, skillful, and sometimes tasteful artistry. That they are of scantily-clothed or nude women is of no consequence; such images have been popular subjects for artists since before the time of "Venus de Milo.” . P10 seeks to enjoin Google from "[ljinking to websites which display or make available PERFECT 10 COPYRIGHTED IMAGES.” Pl.’s Proposed Prelim. Inj. ¶ 1(b). Because "linking" probably does not constitute direct infringement, the only basis for PI0 obtaining this relief would be that such linking infringes vicariously or contributorily. . P10 argues that merely by viewing such websites, individual users of Google search make local "cache” copies of its photos and thereby directly infringe through reproduction. The Court rejects this argument. Local browser caching basically consists of a viewer's computer storing automatically the most recently viewed content of the websites the viewer has visited. It is an automatic process of which most users are unaware, and its use likely is "fair” under 17 U.S.C. § 107 . But cf. Intellectual Reserve, Inc. v. Utah Lighthouse Ministry, Inc., 75 F.Supp.2d 1290 (D.Utah 1999). Local caching by the browsers of individual users is noncommercial, transforma-tive, and no more than necessary to achieve the objectives of decreasing network latency and minimizing unnecessary bandwidth usage (essential to the internet). It has a minimal impact on the potential market for the original work, especially given that most users would not be able to find their own local browser cache, let alone locate a specific cached copy of a particular image. That local browser caching is fair use is supported by a recent decision holding that Google’s own cache constitutes fair use. Field v. Google, Inc., 412 F.Supp.2d 1106 (D.Nev.2006). If anything, the argument that local browser caching is fair use is even stronger. Whereas Google is a commercial entity, individual users are typically noncommercial. Whereas Google arranges to maintain its own cache, individual users typically are not aware that their browsers automatically cache viewed content. Whereas Google's cache is open to the world, an individual’s local browser cache is accessible on that computer alone. . Google also contends that it qualifies for protection under each of the four DMCA safe harbors, 17 U.S.C. § 512 (a)-(d). In light of the ensuing analysis concluding that Google is neither vicariously nor contributorily liable, it is unnecessary for the Court to deal with the DMCA issues. . Cf. Nimmer § 12B.04[A][1], discussing the provision in § 512(d)(l)(A)-(B) of the DMCA that someone cannot qualify for a “safe harbor” if he ignores what has been referred to as a "red flag” of obvious infringement: Given “how difficult it can be to determine whether all the elements of infringement are present'—from proper ownership and standing to lack of license ... to satisfaction of notice formalities ... to the perennially murky issue of fair use, and beyond[, i]t would seem, therefore, that the ‘flag’ must be brightly red indeed—and be waving blatantly in the provider's face—to serve the statutory goal of making ‘infringing activity ... apparent.’ " . See https://www.google.com/adsense/policies (last visited January 27, 2006). . Apparently, Dr. Zada or one of his P10 employees "copied and pasted” the truncated URL that Google displays in its upper frame, rather than taking the extra step of identifying the full file name of the particular infringing image. . These two inquiries are also considered when a defendant seeks the protection of the DMCA § 512(d) (information location tools) safe harbor, which requires that the defendant "not receive a financial benefit directly attributable to the infringing activity [where] the [the defendant] has the right and ability to control such activity....” 17 U.S.C. § 512 (d)(2). . P10 submitted total revenue figures for Google generally and for Google’s AdSense program in particular. It did not, however, attempt to quantify the amount of revenue Google derives from infringing content. . PlO's reliance on Netcom is similarly misplaced. The ability to delete infringing postings to an electronic bulletin board system was held to constitute the ability to control infringing activity on that system, but the "environment” at issue was a closed system. . P10 nevertheless claims that Google can prevent its web crawler from indexing websites with a "history” of infringement and that it can block access to images based on their "verbal” context {i.e., blocking all images that would be responsive to a particular search query). Both measures suffer from the same flaws: imprecision and overbreadth. PIO has not explained how Google could evaluate whether a certain site has a "history” of infringement. For Google to block certain images altogether would suppress many search results that do not infringe. Thus, to filter all images responsive to the search query "Vibe Sorenson” {see attached Exhibit A) would suppress both infringing and noninfringing {e.g., licensed) images. Similarly, because websites are often comprised of dozens, if not hundreds of pages, suppressing an entire website due to the infringing content found on one or more specific pages would result in the suppression of speech and would be against the public interest. . Docket No. 22.