In this аppeal, we consider a copyright owner’s efforts to stop an Internet search engine from facilitating access to infringing images. Perfect 10, Inc. sued Google Inc., for infringing Perfect 10’s copyrighted photographs of nude models, among other claims. Perfect 10 brought a similar action against Amazon.com and its subsidiary A9.com (collectively, “Amazon.com”). The district court preliminarily enjoined Google from creating and publicly displaying thumbnail versions of Perfect 10’s images,
Perfect 10 v. Google, Inc.,
The district court handled this complex case in a particularly thoughtful and skillful manner. Nonetheless, the district court erred on certain issues, as we will further explain below. We affirm in part, reverse in part, and remand.
*711 I
Background
Google’s computers, along with millions of others, are connected to networks known collectively as the “Internet.” “The Internet is a world-wide network of networks ... all sharing a common communications technology.”
Religious Tech. Ctr. v. Netcom On-Line Commc’n Servs., Inc.,
Google operates a search engine, a software program that automatically accesses thousands of websites (collections of web-pages) and indexes them within a database stored on Google’s computers. When a Google user accesses the Google website and types in a search query, Google’s software searches its database for websites responsive to that search query. Google then sends relevant information from its index of websites to the user’s computer. Google’s search engines can provide results in the form of text, images, or videos.
The Google search engine that provides responses in the form of images is called “Google Image Search.” In response to a search query, Google Image Search identifies text in its database responsive to the query and then communicates to users the images associated with the relevant text. Google’s software cannot recognize and index the images themselves. Google Image Search provides search results as a web-page of small images called “thumbnails,” which are stored in Google’s servers. Thе thumbnail images are reduced, lower-resolution versions of full-sized images stored on third-party computers.
When a user clicks on a thumbnail image, the user’s browser program interprets HTML instructions on Google’s webpage. These HTML instructions direct the user’s browser to cause a rectangular area (a “window”) to appear on the user’s computer screen. The window has two separate areas of information. The browser fills the top section of the screen with information from the Google webpage, including the thumbnail image and text. The HTML instructions also give the user’s browser the address of the website publisher’s computer that stores the full-size version of the thumbnail.
2
By following the HTML instructions to access the third-party webpage, the user’s browser connects to the website publisher’s computer, downloads the full-size image, and makes the image appear at the bottom of the window on the user’s screen. Google does not store the images that fill this lower part of the window and does not communicate the images to the user; Google simply provides HTML instructions directing a user’s browser to access a third-party website. However, the top part of the window (containing the information from the Goo
*712
gle webpage) appears to frame and comment on the bottom part of the window. Thus, the user’s window appears to be filled with a single integrated presentation of the full-size image, but it is actually an image from a third-party website framed by information from Google’s website. The process by which the webpage directs a user’s brоwser to incorporate content from different computers into a single window is referred to as “in-line linking.”
Kelly v. Arriba Soft Corp.,
Google also stores webpage content in its cache. 3 For each cached webpage, Google’s cache contains the text of the webpage as it appeared at the time Google indexed the page, but does not store images from the webpage. Id. at 833. Google may provide a link to a cached web-page in response to a user’s search query. However, Google’s cache version of the webpage is not automatically updated when the webpage is revised by its owner. So if the webpage owner updates its web-page to remove the HTML instructions for finding an infringing image, a browser communicating directly with the webpage would not be able to access that image. However, Google’s cache copy of the web-page would still have the old HTML instructions for the infringing image. Unless the owner of the computer changed the HTML address of the infringing image, or otherwise rendered the image unavailable, a browser accessing Google’s cache copy of the website could still access the image where it is stored on the website publisher’s computer. In other words, Google’s cache copy could provide a user’s browser with valid directions to an infringing image even though the updated web-page no longer includes that infringing image.
In additiоn to its search engine operations, Google generates revenue through a business program called “AdSense.” Under this program, the owner of a website can register with Google to become an AdSense “partner.” The website owner then places HTML instructions on its web-pages that signal Google’s server to place advertising on the webpages that is relevant to the webpages’ content. Google’s computer program selects the advertising automatically by means of an algorithm. AdSense participants agree to share the revenues that flow from such advertising with Google.
Google also generated revenues through an agreement with Amazon.com that allowed Amazon.com to in-line link to Google’s search results. Amazon.com gave its users the impression that Amazon.com was providing search results, but Google communicated the search results directly to Amazon.com’s users. Amazon.com routed users’ search queries to Google and automatically transmitted Google’s responses (i.e., HTML instructions for linking to Google’s search results) back to its users.
*713 Perfect 10 markets and sells copyrighted images of nude models. Among other enterprises, it operates a subscription website on the Internet. Subscribers pay a monthly fee to view Perfect 10 images in a “members’ area” of the site. Subscribers must use a password to log into the members’ area. Google does not include these password-protected images from the members’ area in Google’s index or database. Perfect 10 has also licensed Fonestarz Media Limited to sell and distribute Perfect 10’s reduced-size copyrighted images for download and use on cell phones.
Some website publishers republish Perfect 10’s images on the Internet without authorization. Once this occurs, Google’s search engine may automatically index the webpages containing these images and provide thumbnail versions of images in response to user inquiries. When a user clicks on the thumbnail image returned by Google’s search engine, the user’s browser accesses the third-party webpage and inline links to the full-sized infringing image stored on the website publisher’s computer. This image appears, in its original context, on the lower portion of the window on the user’s computer screen framed by information from Google’s webpage.
Procedural History. In May 2001, Perfect 10 began notifying Google that its thumbnail images and in-line linking to the full-size images infringed Perfect 10’s copyright. Perfect 10 continued to send these notices through 2005.
On November 19, 2004, Perfect 10 filed an action against Google that included copyright infringement claims. This was followed by a similar action against Amazon.com on June 29, 2005. On July 1, 2005 and August 24, 2005, Perfect 10 sought a preliminary injunction to prevent Amazon.com and Google, respectively, from “copying, reproducing, distributing, publicly displaying, adapting or otherwise infringing, or contributing to the infringement” of Perfect 10’s photographs; “linking to websites that provide full-size infringing versions of Perfect 10’s photographs; and infringing Perfect 10’s username/password combinations.”
The district court consolidated the two actions and heard both preliminary injunction motions on November 7, 2005. The district court issued orders granting in part and denying in part the preliminary injunction against Google and denying the preliminary injunction against Amazon.com. Perfect 10 and Google cross-appealed the partial grant and partial denial of the preliminary injunction motion, and Perfect 10 appealed the denial of the preliminary injunction against Amazon.com. On June 15, 2006, the district court temporarily stayed the preliminary injunction.
II
Standard of Review
We review the district court’s grant or denial of a preliminary injunction for an abuse of discretion.
A&M Records, Inc. v. Napster, Inc.,
Section 502(a) of the Copyright Act authorizes a court to grant injunctive relief “on such terms as it may deem reasonable to prevent or restrain infringement of a copyright.” 17 U.S.C. § 502(a). “Preliminary injunctive relief is available to a party who demonstrates 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
*714
balance of hardships tips in its favor. These two formulations represent two points on a sliding scale in which the required degree of irreparable harm increases as the probability of success decreases.”
Napster,
Because Perfect 10 has the burden of showing a likelihood of success on the merits, the district court held that Perfect 10 also had the burden of demonstrating a likelihood of overcoming Google’s fair use defense under 17 U.S.C. § 107.
Perfect 10,
However, entitlement for preliminary relief “is determined in the context of the presumptions and burdens that would inhere at trial on the merits.”
H.H. Robertson,
Google also raises an affirmative defense under title II of the Digital Millennium Copyright Act (“DMCA”), 17 U.S.C. § 512. Congress enacted title II of the DMCA “to provide greater certainty to service providers concerning their legal exposure for infringements that may occur in the course of them activities.”
Ellison v. Robertson,
Ill
Direct Infringement
Perfect 10 claims that Google’s search engine program directly infringes two exclusive rights granted to copyright holders: its display rights and its distribution rights.
5
“Plaintiffs must satisfy two requirements to present a prima facie case of direct infringement: (1) they must show ownership of the allegedly infringed material and (2) they must demonstrate that the alleged infringers violate at least one exclusive right granted to copyright holders under 17 U.S.C. § 106.”
Napster,
Perfect 10’s ownership of at least some of the images at issue is not disputed.
See Perfect 10,
The district court held that Perfect 10 was likely to prevail in its claim that Google violated Perfect 10’s display right with respect to the infringing thumbnails.
Id.
at 844. However, the district court concluded that Perfect 10 was not likely to prevail on its claim that Google violated either Perfect 10’s display or distribution right with respect to its full-size infringing images.
Id.
at 844-45. We review these rulings for an abuse of discretion.
Napster,
A. Display Right
In considering whether Perfect 10 made a prima facie case of violation of its display right, the district court reasoned that a computer owner that stores an image as electronic information and serves that electronic information directly to the user (“i.e., physically sending ones and zeroes
*716
over the [I]nternet to the user’s browser,”
Perfect 10,
Applying the server test, the district court concluded that Perfect 10 was likely to succeed in its claim that Google’s thumbnails constituted direct infringement but was unlikely to succeed in its claim that Google’s in-line linking to full-size infringing images constituted a direct infringement. Id. at 843-45. As explained below, because this analysis comports with the language of the Copyright Act, we agree with the district court’s resolution of both these issues.
We have not рreviously addressed the question when a computer displays a copyrighted work for purposes of section 106(5). Section 106(5) states that a copyright owner has the exclusive right “to display the copyrighted work publicly.” The Copyright Act explains that “display” means “to show a copy of it, either directly or by means of a film, slide, television image, or any other device or process.... ” 17 U.S.C. § 101. Section 101 defines “copies” as “material objects, other than pho-norecords, in which a work is fixed by any method now known or later developed, and from which the work can be perceived, reproduced, or otherwise communicated, either directly or with the aid of a machine or device.” Id. Finally, the Copyright Act provides that “[a] work is ‘fixed’ in a tangible medium of expression when its embodiment in a copy or phonorecord, by or under the authority of the author, is sufficiently permanent or stable to permit it to be perceived, reproduced, or otherwise communicated for a period of more than transitory duration.” Id.
We must now apply these definitions to the facts of this case. A photographic image is a work that is “ ‘fixed’ in a tangible medium of expression,” for purposes of the Copyright Act, when embodied (i.e., stored) in a computer’s server (or hard disk, or other storage device). The image stored in the computer is the “copy” of the work for purposes of copyright law.
See MAI Sys. Corp. v. Peak Computer, Inc.,
Google does not, however, display a copy of full-size infringing photographic images for purposes of the Copyright Act when Google frames in-line linked images that appear on a user’s computer screen. Because Google’s computers do not store the photographic images, Google does not have a copy of the images for purposes of the Copyright Act. In other words, Google does not have any “material objects ... in which a work is fixed ... and from which the work can be perceived, reproduced, or otherwise communicated” and thus cannot communicate a copy. 17 U.S.C. § 101.
Instead of communicating a copy of the image, Google provides HTML instructions that direct a user’s browser to a website publisher’s computer that stores the full-size photographic image. Providing these HTML instructions is not equivalent to showing a
copy.
First, the HTML instructions are lines of text, not a photographic image. Second, HTML instructions do not themselves cause
infringing
images to appear on the user’s computer screen. The HTML merely gives the address of the image to the user’s browser. The browser then interacts with the computer that stores the infringing image. It is this interaction that causes an infringing image to appear on the user’s computer screen. Google may facilitate the user’s access to infringing images. However, such assistance raises only contributory liability issues,
see Metro-Goldwyn-Mayer Studios, Inc. v. Grokster, Ltd.,
Perfect 10 argues that Google displays a copy of the full-size images by framing the full-size images, which gives the impression that Google is showing the image within a single Google webpage. While in-line linking and framing may cause some computer users to believe they are viewing a single Google webpage, the Copyright Act, unlike the Trademark Act, does not protect a copyright holder against acts that cause consumer сonfusion. Cf. 15 U.S.C. § 1114(l)(providing that a person who uses a trademark in a manner likely to cause confusion shall be liable in a civil action to the trademark registrant). 7
Nor does our ruling that a computer owner does not display a copy of an image when it communicates only the HTML address of the copy erroneously collapse the display right in section 106(5) into the reproduction right set forth in section 106(1). Nothing in the Copyright
*718
Act prevents the various rights protected in section 106 from overlapping. Indeed, under some circumstances, more than one right must be infringed in order for an infringement claim to arise. For example, a “Game Genie” device that allowed a player to alter features of a Nintendo computer game did not infringe Nintendo’s right to prepare derivative works because the Game Genie did not incorporate any portion of the game itself.
See Lewis Galoob Toys, Inc. v. Nintendo of Am., Inc.,
Because Google’s cache merely stores the text of webpages, our analysis of whether Google’s search engine program potentially infringes Perfect 10’s display and distribution rights is equally applicable to Google’s cache. Perfect 10 is not likely to succeed in showing that a сached webpage that in-line links to full-size infringing images violates such rights. For purposes of this analysis, it is irrelevant whether cache copies direct a user’s browser to third-party images that are no longer available on the third party’s website, because it is the website publisher’s computer, rather than Google’s computer, that stores and displays the infringing image.
B. Distribution Right
The district court also concluded that Perfect 10 would not likely prevail on its claim that Google directly infringed Perfect 10’s right to distribute its full-size images.
Perfect 10,
Again, the district court’s conclusion on this point is consistent with the language of the Copyright Act. Section 106(3) provides that the copyright owner has the exclusive right “to distribute copies or pho-norecords of the copyrighted work to the public by sale or other transfer of ownership, or by rental, lease, or lending.” 17 U.S.C. § 106(3). As noted, “copies” means “material objects ... in which a work is fixed.” 17 U.S.C. § 101. The Supreme Court has indicated that in the electronic context, copies may be distributed electronically.
See N.Y. Times Co. v. Tasini,
Perfect 10 incorrectly relies on
Hotaling v. Church of Jesus Christ of Latter-Day Saints
and
Napster
for the proposition that merely making images “available” violates the copyright owner’s distribution right.
Hotaling v. Church of Jesus Christ of Latter-Day Saints,
This “deemed distribution” rule does not apply to Google. Unlike the participants in the Napster system or the library in Hotaling, Google does not own a collection of Perfect 10’s full-size images and does not communicate these images to the computers of people using Google’s search engine. Though Google indexes these images, it does not have a collection of stored full-size images it makes available to the public. Google therefore cannot be deemed to distribute copies of these images under the reasoning of Napster or Hotaling. Accordingly, the district court correctly concluded that Perfect 10 does not have a likelihood of success in proving that Google violates Perfect 10’s distribution rights with respect to full-size images.
C. Fair Use Defense
Although Perfect 10 has succeeded in showing it would prevail in its prima facie case that Google’s thumbnail images infringe Perfect 10’s display rights, Perfect 10 must still show a likelihood that it will prevail against Google’s affirmative defense. Google contends that its use of thumbnails is a fair use of the images and therefore does not constitute an infringement of Perfect 10’s copyright. See 17 U.S.C. § 107.
The fair use defense permits the use of copyrighted works without the copyright owner’s consent under certain situations. The defense encourages and allows the development of new ideas that build on earlier ones, thus providing a necessary counterbalance to the copyright law’s goal of protecting creators’ work product. “From the infancy of copyright protection, some opportunity for fair use of copyrighted materials has been thought necessary to fulfill copyright’s very purpose ....”
Campbell,
Congress codified the common law of fair use in 17 U.S.C. § 107, which provides:
Notwithstanding the provisions of sections 106 and 106A, the fair use of a copyrighted work, including such use by reproduction in copies or phonorecords or by any other means specified by that section, for purposes such as criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research, is not an infringement of copyright. In determining whether the use made of a work in any particular case is a fair use the factors to be considered shall include—
(1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
(2) the nature of the copyrighted work;
(3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
*720 (4) the effect of the use upon the potential market for or value of the copyrighted work.
The fact that a work is unpublished shall not itself bar a finding of fair use if such finding is made upon consideration of all the above factors.
17 U.S.C. § 107.
We must be flexible in applying a fair use analysis; it “is not to be simplified with bright-line rules, for the statute, like the doctrine it recognizes, calls for case-by-case analysis.... Nor may the four statutory factors be treated in isolation, one from another. All are to be explored, and the results weighed together, in light of the purposes of copyright.”
Campbell,
In applying the fair use analysis in this case, we are guided by
Kelly v. Arriba Soft Corp.,
which considered substantially the same use of copyrighted photographic images as is at issue here.
See
In this case, the district court determined that Google’s use of thumbnails was not a fair use and distinguished
Kelly. Perfect 10,
Purpose and character of the use.
The first factor, 17 U.S.C. § 107(1), requires a court to consider “the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes.” The central purpose of this inquiry is to determine whether and to what extent the new work is “transformative.”
Campbell,
*721
As noted in
Campbell,
a “transformative work” is one that alters the original work “with new expression, meaning, or message.”
Campbell,
Google’s use of thumbnails is highly transformative. In
Kelly,
we concluded that Arriba’s use of thumbnails was transformative because “Arriba’s use of the images serve[d] a different function than Kelly’s use—improving access to information on the [I]nternet versus artistic expression.”
Kelly,
The fact that Google incorporates the entire Perfect 10 image into the search engine results does not diminish the trans-formative nature of Google’s use. As the district court correctly noted,
Perfect 10,
*722
copy serves a different function than the original work,
Kelly,
The district court nevertheless determined that Google’s use of thumbnail images was less transformative than Arriba’s use of thumbnails in
Kelly
because Google’s use of thumbnails superseded Perfect 10’s right to sell its reduced-size images for use on cell phones.
See Perfect 10,
Additionally, the district court determined that the commercial nature of Google’s use weighed against its transforma-tive nature.
Id.
Although
Kelly
held that the commercial use of the photographer’s images by Arriba’s search engine was less exploitative than typical commercial use, and thus weighed only slightly against a finding of fair use,
Kelly,
In conducting our case-specific analysis of fair use in light of the purposes of copyright,
Campbell,
We note that the superseding use in this case is not significant at present: the district court did not find that any down loads for mobile phone use had taken place.
See Perfect 10,
We conclude that the significantly trans-formative nature of Google’s search engine, particularly in light of its public benefit, outweighs Google’s superseding and commercial uses of the thumbnails in this case. In reaching this conclusion, we note the importance of analyzing fair use flexibly in light of new circumstances.
Sony,
Accordingly, we disagree with the district court’s conclusion that because Google’s use of the thumbnails could supersede Perfect 10’s cell phone download use and because the use was more commercial than Arriba’s, this fair use factor weighed “slightly” in favor of Perfect 10.
Perfect 10,
The nature of the copyrighted work.
With respect to the second factor, “the nature of the copyrighted work,” 17 U.S.C. § 107(2), our decision in
Kelly
is directly on point. There we held that the photographer’s images were “creative in nature” and thus “closer to the core of intended copyright protection than are more fact-based works.”
Kelly,
Here, the district court found that Perfect 10’s images were creative but also previously published.
Peffect 10,
The amount and substantiality of the portion used.
“The third factor asks whether the amount and substantiality of the portion used in relation to the copyrighted work as a whole ... are reasonable in relation to the purpose of the copying.”
Campbell,
Effect of use on the market.
The fourth factor is “the effect of the use upon the potential market for or value of the copyrighted work.” 17 U.S.C. § 107(4). In
Kelly,
we concluded that Arriba’s use of the thumbnail images did not harm the market for the photographer’s full-size images.
See Kelly,
Perfect 10 argues that the district court erred because the likelihood of market harm may be presumed if the intended use of an image is for commercial gain. However, this presumption does not arise when a work is transformative because “market substitution is at least less certain, and market harm may not be so readily inferred.”
Campbell,
Perfect 10 also has a market for reduced-size images, an issue not considered in
Kelly.
The district court held that “Google’s use of thumbnails likely does harm the potential market for the downloading of [Perfect 10’s] reduced-size images onto cell phones.”
Perfect 10,
Having undertaken a case-specific analysis of all four factors, we now weigh these factors together “in light of the purposes of copyright.”
Campbell,
IV
Secondary Liability for Copyright Infringement
We now turn to the district court’s ruling that Google is unlikely to be secondarily liable for its in-line linking to infringing full-size images under the doctrines of contributory and vicarious infringement.
10
The district court ruled that Perfect 10 did not have a likelihood of proving success on the merits of either its contributory infringement or vicarious infringement claims with respect to the full-size images.
See Perfect 10,
Direct Infringement by Third Parties.
As a threshold matter, before we examine Perfect 10’s claims that Google is secondarily liable, Perfect 10 must establish that there has been direct infringement by third parties.
See Napster,
Perfect 10 alleges that third parties directly infringed its images in three ways. First, Perfect 10 claims that third-party *726 websites directly infringed its copyright by reproducing, displaying, and distributing unauthorized copies of Perfect 10’s images. Google does not dispute this claim on appeal.
Second, Perfect 10 claims that individual users of Google’s search engine directly infringed Perfect 10’s copyrights by storing full-size infringing images on their computers. We agree with the district court’s conclusion that Perfect 10 failed to provide sufficient evidence to support this claim.
See Perfect 10,
Finally, Perfect 10 contends that users who link to infringing websites automatically make “cache” copies of full size images and thereby directly infringe Perfect 10’s reproduction right. The district court rejected this argument, holding that any such reproduction was likely a “fair use.” Id. at 852 n. 17. The district court reasoned that “[l]ocal caching by the browsers of individual users is noncommercial, transformative, аnd no more than necessary to achieve the objectives of decreasing network latency and minimizing unnecessary bandwidth usage (essential to the [I]nternet). It has a minimal impact on the potential market for the original work.... ” Id. We agree; even assuming such automatic copying could constitute direct infringement, it is a fair use in this context. The copying function performed automatically by a user’s computer to assist in accessing the Internet is a transfor-mative use. Moreover, as noted by the district court, a cache copies no more than is necessary to assist the user in Internet use. It is designed to enhance an individual’s computer use, not to supersede the copyright holders’ exploitation of their works. Such automatic background copying has no more than a minimal effect on Perfect 10’s rights, but a considerable public benefit. Because the four fair use factors weigh in favor of concluding that cache copying constitutes a fair use, Perfect 10 has not carried its burden of showing that users’ cache copies of Perfect 10’s full-size images constitute direct infringement.
Therefore, we must assess Perfect 10’s arguments that Google is secondarily liable in light of the direct infringement that is undisputed by the parties: third-party websites’ reproducing, displaying, and distributing unauthorized copies of Perfect 10’s images on the Internet. Id. at 852.
A. Contributory Infringement
In order for Perfect 10 to show it will likely succeed in its contributory liability claim against Google, it must establish that Google’s activities meet the definition of contributory liability recently enunciated in
Grokster.
Within the general rule thаt “[o]ne infringes contributorily by intentionally inducing or encouraging direct infringement,”
Grokster,
Looking at the second category of liability identified by the Supreme Court (distributing products), Google relies on
Sony,
*727
We must next consider whether Google could be held liable under the first category of contributory liability identified by the Supreme Court, that is, the liability that may be imposed for intentionally encouraging infringement through specific acts.
11
Grokster
tells us that contribution to infringement must be intentional for liability to arise.
Grokster,
Our tests for contributory liability are consistent with the rule set forth in
Grok-ster.
We have adopted the general rule set, forth in
Gershwin Publishing Corp. v. Columbia Artists Management, Inc.,
namely: “one who, with knowledge of the infringing activity, induces, causes or matеrially contributes to the infringing conduct of another, may be held liable as a ‘contributory’ infringer,”
We have further refined this test in the context of cyberspace
12
to determine when contributory liability can be imposed on a provider of Internet access or services.
See Napster,
The
Napster
test for contributory liability was modeled on the influential district court decision in
Religious Technology Center v. Netcom On-Line Communication Services, Inc. (Netcom),
Although neither Napster
nor
Netcom
expressly required a finding of intent, those cases are consistent with
Grokster
because both decisions ruled that a service provider’s knowing failure to prevent infringing actions could be the basis for imposing contributory liability. Under such circumstances, intent may be imputed. In addition,
Napster
and
Netcom
are consistent with the longstanding requirement that an actor’s contribution to infringement must be material to warrant the imposition of contributory liability.
Gershwin,
Here, the district court held that even assuming Google had actual knowledge of infringing material available on its system, Google did not materially contribute to infringing conduct because it did not undertake any substantial promotional or advertising efforts to encourage visits to infringing websites, nor provide a significant revenue stream to the infringing websites.
Perfect 10,
The district court did not resolve the factual disputes over the adequacy of Perfect 10’s notices to Google and Google’s responses to these notices. Moreover, there are factual disputes over whether there are reasonable and feasible means for Google to refrain from providing access to infringing images. Therefore, we must remand this claim to the district court for further consideration whether Perfect 10 would likely succeed in establishing that Google was contributorily liable for in-line linking to full-size infringing images under the test enunciated today. 13
B. Vicarious Infringement
Perfect 10 also challenges the district court’s conclusion that it is not likely to prevail on a theory of vicarious liability against Google.
Perfect 10,
We evaluate Perfect 10’s arguments that Google is vicariously liable in light of the direct infringement that is undisputed by the parties, namely, the third-party websites’ reproduction, display, and distribution of unauthorized copies of Perfect 10’s images on the Internet.
Perfect 10,
With respect to the “control” element set forth in
Grokster,
Perfect 10 has not demonstrated a likelihood of showing that Google has the legal right to stop or limit the direct infringement of third-party websites.
See Grokster,
Nor is Google similarly situated to Napster. Napster users infringed the plaintiffs’ reproduction and distribution rights through their use of Napster’s proprietary music-file sharing system.
Napster,
Moreover, the district court found that Google lacks the practical ability to police the third-party websites’ infringing conduct.
Id.
at 857-58. Specifically, the court found that Google’s supervisory power is limited because “Google’s software lacks the ability to analyze every image on the [I]nternet, compare each image to all the other copyrighted images that exist in the world ... and determine whether a certain image on the web infringes someone’s copyright.”
Id.
at 858. The district court also concluded that Perfect 10’s suggestions regarding measures Google could implement to prevent its web crawler from indexing infringing websites and to block access to infringing images were not workable.
Id.
at 858 n. 25. Rather, the suggestions suffered from both “imprecision and overbreadth.”
Id.
We hold that these findings are not clearly erroneous. Without image-recognition technology, Google lacks the practical ability to police the infringing activities of third-party websites. This distinguishes Google from the defendants held liable in
Napster
and
Fo-novisa. See Napster,
Perfect 10 argues that Google could manage its own operations to avoid indexing websites with infringing content and linking to third-party infringing sites. This is a claim of contributory liability, not vicarious liability. Although “the lines between direct infringement, contributory infringement, and vicarious liability are not clearly drawn,”
Sony,
Because we conclude that Perfect 10 has not shown a likelihood of establishing Google’s right and ability to stop or limit the directly infringing conduct of third-party websites, we agree with the district court’s conclusion that Perfect 10 “has not established a likelihood
of
proving the[eontrol] prong necessary for vicarious liability.”
Perfect 10,
*732 C. Digital Millennium Copyright Act
Google claims that it qualifies for the limitations on liability set forth in title II of the DMCA, 17 U.S.C. § 512. In particular, section 512(d) limits the liability of a service provider “for infringement of copyright by reason of the provider referring or linking users to an online location containing infringing material or infringing activity, by using information location tools, including a directory, index, reference, pointer, or hypertext link” if the service provider meets certain criteria. We have held that the limitations on liability contained in 17 U.S.C. § 512 protect secondary infringers as well as direct in-fringers.
Napster,
The parties dispute whether Google meets the specified criteria. Perfect 10 claims that it sent qualifying notices to Google and Google did not act expeditiously to remove the infringing material. Google claims that Perfect 10’s notices did not comply with the notice provisions of section 512 and were not adequate to inform Google of the location of the infringing images on the Internet or identify the underlying copyrighted work. Google also claims that it responded to all notices it received by investigating the webpages identified by Perfect 10 and suppressing links to any webpages that Google confirmed were infringing.
Because the district court determined that Perfect 10 was unlikely to succeed on its contributory and vicarious liability claims, it did not reach Google’s arguments under section 512. In revisiting the question of Perfect 10’s likelihood of success on its contributory infringement claims, the district court should also consider whether Perfect 10 would likely succeed in showing that Google was not entitled to the limitations on injunctive relief provided by title II of the DMCA.
V
Amazon.com
Perfect 10 claims that Amazon.com displays and distributes Perfect 10’s copyrighted images and is also secondarily liable for the infringements of third-party websites and Amazon.com users. The district court concluded that Perfect 10 was unlikely to succeed in proving that Amazon.сom was a direct infringer, because it merely in-line linked to the thumbnails on Google’s servers and to the full-size images on third-party websites. 16 Perfect 10 v. Amazon, No. 05-4753, consolidated with 04-9484 (C.D.Cal. February 21, 2006) (order denying preliminary injunction). In addition, the district court concluded that Perfect 10’s secondary infringement claims against Amazon.com were likely to fail because Amazon.com had no program analogous to AdSense, and thus did not provide any revenues to infringing sites. Id. Finally, the district court determined that Amazon.eom’s right and ability to control the infringing conduct of third-party websites was substantially less than Google’s. Id. Therefore, the district court denied Perfect 10’s motion for a preliminary injunction against Amazon.com. Id.
We agree that Perfect 10 has not shown a likelihood that it would prevail on the merits of its claim that Amazon.com directly infringed its images. Amazon.com communicates to its users only the HTML *733 instructions that direct the users’ browsers to Google’s computers (for thumbnail images) or to a third party’s computer (for full-size infringing images). Therefore, Amazon.com does not display or distribute a copy of the thumbnails or full-size images to its users.
We also agree with the district court’s conclusion that Amazon.com does not have “the right and ability to supervise the infringing activity” of Google or third parties. The district court did not clearly err in concluding that Amazon.com lacked a direct financial interest in such activities. Therefore, Perfect 10’s claim that Amazon.com is vicariously liable for third-party infringement is unlikely to succeed.
However, the district court did not consider whether Amazon.com had
“actual
knowledge that
specific
infringing material is available using its system,” Napster,
VI
We conclude that Perfect 10 is unlikely to succeed in overcoming Google’s fair use defense, and therefore we reverse the district court’s determination that Google’s thumbnail versions of Perfect 10’s images likely constituted a direct infringement. The district court also erred in its secondary liability analysis because it failed to consider whether Google and Amazon.com knew of infringing activities yet failed to take reasonable and feasible steps to refrain from providing access to infringing images. Therefore we must also reverse the district court’s holding that Perfect 10 was unlikely to succeed on the merits of its secondary liability claims. Due to this error, the district court did not consider whether Google and Amazon.com are entitled to the limitations on liability set forth in title II of the DMCA. The question whether Google and Amazon.com are secondarily liable, and whether they can limit that liability pursuant to title II of the DMCA, raise fact-intensive inquiries, potentially requiring further fact finding, and thus can best be resolved by the district court on remand. We therefore remand this matter to the district court for further proceedings consistent with this decision.
Because the district court will need to reconsider the appropriate scope of in-junctive relief after addressing these secondary liability issues, we do not address the parties’ arguments regarding the scope of the injunction issued by the district court. For the same reason, we do not address the parties’ dispute over whether the district court abused its discretion in determining that Perfect 10 satisfied the irreparable harm element of a preliminary injunction.
Therefore, we reverse the district court’s ruling and vacate the preliminary injunction regarding Google’s use of thumbnail versions of Perfect 10’s im *734 ages. 17 We reverse the district court’s rejection of the claims that Google and Amazon.com are secondarily liable for infringement of Perfect 10’s full-size images. We otherwise affirm the rulings of the district court. We remand this matter for further proceedings consistent with this opinion. Each party shall bear its own costs on appeal. See Fed. R. App. P. 39(a)(4).
AFFIRMED IN PART; REVERSED IN PART; REMANDED.
Notes
. Google argues that we lack jurisdiction over the preliminary injunction to the extent it enforces unregistered copyrights. Registration is generally a jurisdictional prerequisite to a suit for copyright infringement.
See
17 U.S.C. § 411. But sеction 411 does not limit the remedies a court can grant. Rather, the Copyright Act gives courts broad authority to issue injunctive relief.
See
17 U.S.C. § 502(a). Once a court has jurisdiction over an action for copyright infringement under section 411, the court may grant injunctive relief to restrain infringement of any copyright, whether registered or unregistered.
See, e.g., Olan Mills, Inc. v. Linn Photo Co.,
. The website publisher may not actually store the photographic images used on its webpages in its own computer, but may provide HTML instructions directing the user’s browser to some further computer that stores the image. Because this distinction does not affect our analysis, for convenience, we will assume that the website publisher stores all images used on its webpages in the website publisher's own computer.
. Generally, a "cache” is "a computer memory with very short access time used for storage of frequently or recently used instructions or data.”
United States v. Ziegler,
. Perfect 10 argues that we are bound by the language and structure of title II of the DMCA in determining Google’s liability for copyright infringement. We have noted that the DMCA does not change copyright law; rather, “Congress provided that [the DMCA's] limitations of liability apply if the provider is found to be liable under existing principles of law.”
Ellison,
. 17 U.S.C. § 106 states, in pertinent part:
Subject to sections 107 through 122, the owner of copyright under this title has the exclusive rights to do and to authorize any of the following:
(1) to reproduce the copyrighted work in copies or phonorecords;
(3) to distribute copies or phonorecords of the copyrighted work to the public by sale or other transfer of ownership, or by rental, lease, or lending;
(5) in the case of literary, musical, dramatic, and choreographic works, pantomimes, and pictorial, graphic, or sculptural works, including the individual images of a motion picture or other audiovisual work, to display the copyrighted work publicly ....
. Because Google initiates and controls the storage and communication of these thumb
*717
nail images, we do not address whether an entity that merely passively owns and manages an Internet bulletin board or similar system violates a copyright owner's display and distribution rights when the users of the bulletin board or similar system post infringing works.
Cf. CoStar Group, Inc. v. LoopNet, Inc.,
. Perfect 10 also argues that Google violates Perfect 10’s right to display full-size images because Google's in-line linking meets the Copyright Act’s definition of "to perform or display a work 'publicly.' " 17 U.S.C. § 101. This phrase means "to transmit or otherwise communicate a performance or display of the work to ... the public, by means of any device or process, whether the members of the public capable of receiving the performance or display receive it in the same place or in separate places and at the same time or at different times.” Id. Perfect 10 is mistaken. Google’s activities do not meet this definition because Google transmits or communicates only an address which directs a user’s browser to the location where a copy of the full-size image is displayed. Google does not communicate a display of the work itself.
. We reject at the outset Perfect 10's argument that providing access to infringing web
*721
sites cannot be deemed transformative and is inherently not fair use. Perfect 10 reliеs on
Video Pipeline, Inc. v. Buena Vista Home Entm't, Inc.,
Unlike the alleged infringers in Video Pipeline and Atari Games, who intentionally misappropriated the copyright owners' works for the purpose of commercial exploitation, Google is operating a comprehensive search engine that only incidentally indexes infringing websites. This incidental impact does not amount to an abuse of the good faith and fair dealing underpinnings of the fair use doctrine. Accordingly, we conclude that Google's inclusion of thumbnail images derived from infringing websites in its Internet-wide search engine activities does not preclude Google from raising a fair use defense.
. Google contends that Perfect 10’s photographic images are less creative and less deserving of protection than the images of the American West in
Kelly
because Perfect 10 boasts of its un-retouched photos showing the natural beauty of its models. Having reviewed the record, we conclude that the district court’s finding that Perfect 10’s photographs "consistently reflect professional, skillful, and sometimes tasteful artistry” is not clearly erroneous.
Perfect 10,
. Because the district court concluded that Perfect 10 was likely to prevail on its direct infringement claim with respect to Google’s use of thumbnails, but not with respect to its in-line linking to full-size images, the district court considered Google’s potential secondary liability only on the second issue.
. Google’s activities do not meet the "inducement” test explained in
Grokster
because Google has not promoted the use of its search engine specifically to infringe copyrights.
See Grokster,
. “Cyberspace is a popular term for the world of electronic communications over computer networks.”
Religious Tech. Ctr. v. Netcom On-Line Commc’n Servs., Inc.,
. Perfect 10 claims that Google materially contributed to infringement by linking to websites containing unauthorized passwords, which enabled Google users to access Perfect 10's website and make infringing copies of images. However, Perfect 10 points to no evidence that users logging onto the Perfect 10 site with unauthorized passwords infringed Perfect 10’s exclusive rights under section 106. In the absence of evidence that Google's actions led to any direct infringement, this argument does not assist Perfect 10 in establishing that it would prevail on the merits of its contributory liability claim.
See Napster,
. Napster’s system included "Napster's Mu-sicShare software, available free of charge from Napster's Internet site, and Napster’s network servers and server-side software.”
Napster,
. Having so concluded, we need not reach Perfect 10’s argument that Google received a direct financial benefit.
. Amazon.com states that it ended its relationship with Google on April 30, 2006. Perfect 10’s action for preliminary injunction against Amazon.com is not moot, however, because Amazon.com has not established "that the allegedly wrongful behavior cannot reasonably be expected to recur.”
F.T.C. v. Affordable Media, LLC,
. Because we vacate the injunction, Google's motion for stay of the injunction is moot.
