Lehrman v. Lovo, Inc., — F.Supp.3d —-, 2025 WL 1902547,
24-CV-3770 (JPO) (S.D.N.Y. Jul. 10, 2025)
Plaintiffs alleged that Lovo misled them about its use of
their voices, using AI to synthesize and sell unauthorized “clones” of their
voices. They sued for violations of New York civil rights and consumer
protection laws, the Lanham Act, and the Copyright Act, along with common-law
contract, fraud, conversion, unjust enrichment, and unfair competition claims.
Lovo was partially successful on its motion to dismiss, but had to deal with
the NY ROP claims, as well as consumer protection and breach of contract
claims. I might not agree with every detail of the extensive opinion,
but this is a useful map of the issues. It’s also a reminder of the hazards of
not contracting properly with people you don’t employ.
Plaintiffs Lehrman and Sage are voice-over actors. Lovo
“sells a text-to-speech subscription service that allows its clients to
generate voice-over narrations at a fraction of the cost of the traditional
model.” It claims that its software, Genny, was “created using ‘1000s of
voices.’ ” Genny is allegedly capable of creating a voice “clone,” which
“refers to a virtual copy of a real person’s voice.” It advertised its voice
cloning service by emphasizing how similar its cloned voices are to the
originals from which they are derived. In one marketing video, Lovo stated,
“you will hear five speakers whose voices have been cloned to near perfection.
Their tone, accent, and even mannerisms are fully learned by our AI voice
system.” One of Lovo’s co-founders described a cloned voice as a replacement
for “a real human voice,” allowing users of Lovo’s platform to “make that voice
say anything that you want, even if that person has never actually said that
before in their life.”
Lovo advertises the commercial use of its platform to “Save
$$ and time on voiceovers.” Allegedly, “Lovo represents to its customers that
Lovo is granting full commercial rights for all content generated using its
platform to users who subscribe to any of its paid plans,” which cover “any
monetized, business-related uses such as videos, audio books, advertising,
promotion, web page blogging, [and] product integration.”
Lovo originally solicited Lehrman and Sage for the projects
relevant to this action on an online marketplace for freelance services called
Fiverr. A Lovo employee allegedly contacted Lehrman and hired him to “provide
voice recordings for ‘research purposes,’” assuring him that the company would
use the recordings for “research purposes only,” and that such research would
be only “internal” and “academic” in nature. Lehrman wrote back, asking for a
“guarantee that these scripts will not be used for anything other than your
specific research project,” to which the employee replied: “The scripts will
not be used for anything else ….” The employee further confirmed that the
script would not be “repurposed and used in a different order.” He was paid
$1,200 for his work.
Similarly, Sage asked what the proposed recordings would be
used for, and the Lovo rep (a principal) replied: “These are test scripts for
radio ads. They will not be disclosed externally, and will only be consumed
internally, so will not require rights of any sort.” Sage asked him to to
confirm that the recordings would not be used “in broadcast,” and he again
repeated the previous statement. Lovo paid Sage $400 for her work.
Both of them ultimately (after filing suit) registered their
scripted performances with the Copyright Office.
Plaintiffs allegedly first learned that their voices had
been used in unanticipated ways when they listened to an episode of the
Deadline Strike Talk podcast narrated in part by an artificial voice produced
by Lovo’s software. They alleged that the voice used in the podcast was
“identical to Mr. Lehrman’s voice.” Allegedly, “[n]umerous people who heard the
podcast,” including “friends” and “professional colleagues” “told Mr. Lehrman
or Ms. Sage” that the “voice on the podcast was virtually identical to Mr. Lehrman’s
voice,” and that “the cloned voice would undoubtedly be mistaken for Mr.
Lehrman’s actual voice.” They also alleged, with declaration support, that
professionals “experience[d] [in] discerning and conveying small differences in
voice tone, quality, timbre, and delivery” believe that “Lovo’s cloned voice is
a replica of Mr. Lehrman’s real voice.” Lehrman allegedly found that Lovo “had
been marketing [the clone of his voice] as part of its subscription service
under the stage name ‘Kyle Snow.’ ” The Kyle Snow voice was also in
advertisements on Lovo’s website and YouTube. For example, Lovo allegedly
advertised the Kyle Snow voice as “an ideal male voice generator … for all
kinds of content” due to his “upbeat tone and slightly faster talking speed.”
Sage also allegedly discovered that Lovo had created a clone
of her voice named “Sally Coleman” that was available to Lovo’s subscribers.
This was allegedly marketed using “side-by-side” comparisons of Sage’s original
audio recordings—the ones she provided via Fiverr—and the “cloned version of
her voice,” including in an investor pitch that was posted on YouTube.
Plaintiffs brought individual claims and also sought to
represent a voice actor class and a consumer class of consumers who bought the
Lovo software and used the voices.
Breach of contract: Sufficiently pled.
Lanham Act false association: Failed for want of a mark that
was used as a mark. Plaintiffs didn’t argue that they could bring §43(a)(1)(A)
claims without a valid mark. Persona can be a “mark”—even for a noncelebrity.
But celebrity endorsement cases have involved “the unauthorized use of a
celebrity’s likeness on merchandise, implying that the celebrity approves of
the merchandise or is affiliated with the seller,” or “the explicit use of a
celebrity’s likeness in advertising or to promote goods or services.”
Plaintiffs alleged that “they are well-known and
sought-after voice actors whose voices are their recognizable calling cards,”
which was enough at this stage to make the use of their voices capable of
causing consumer confusion. The problem was that the allegations didn’t suggest
trademark use. “Because marks can take essentially any form, courts must
therefore be careful to ensure that they receive protection only when used as
contemplated by the statute—that is, as marks.” The court found celebrity
endorsement cases to be an “uneas[y]” fit with this principle, “as celebrities’
personas are also their products,” though at least for advertising cases the
service of endorsement seems like the trademark use. Citing Jennifer Rothman,
the court emphasized that “personal marks” are treated differently from other
marks—they’re harder to register, and the law “is highly skeptical of efforts
to restrict individuals from using their own identities in trade” and doesn’t
allow “transfer attributes of one’s identity—such as personal skill as an
artisan—when one transfers trademark rights in one’s name or likeness.” [My own
view is that the latter point isn’t really relevant—it’s just not possible to
do that; maybe that means that many name transfers are transfers in gross and
should fail, but that is straying pretty far from our concerns here.]
The key issue here was that this wasn’t a celebrity
endorsement case: the voices were themselves the products/services being sold.
“Plaintiffs here use their voices in ways that are clearly separable from their
identities and personalities. Their clients pay them to produce recordings of
themselves narrating scripts, which the clients then own and use to produce
content, as authorized by their contracts with Plaintiffs.” Thus, their voices
“serve dual functions as both ‘one of the most palpable ways identity is
manifested,’ and as ordinary services in the voice-over market.” Thus, even as
to the alleged false endorsement or “business affiliation” confusion,
plaintiffs’ voices were “protectable only to the extent that they function
primarily as source identifiers rather than as products themselves.” The court
analogized to trade dress product design claims, which are difficult because trademark
terminology “is unsuited for application to the product itself.”
Plaintiffs failed to show that their voices were protectable
marks. They didn’t plead secondary meaning or the relevant factors except for “a
few conclusory references to the recognizability of their voices” and the fact
that “Plaintiffs’ work has been sought out by large companies.” Regardless, “Plaintiffs
have not alleged that their voices are primarily significant as brands rather
than as services to which brands might be attached.” As the court noted, “even
extremely famous celebrities are barred from asserting Section 43(a)(1) claims
based on the use of their likenesses as products rather than as
source-identifying marks.” Product use was descriptive use, not trademark use.
An alternate rule would threaten a new voice actor “whose voice happens to
sound highly similar to either Lehrman’s or Sage’s,” especially given that
identicality isn’t required for likely confusion. The court also pointed out
that unregistered trademark rights can be assigned, with “unsettling”
consequences for voices:
To allow any artist, actor, or
other creative tradesperson to sue their doppelgangers for trademark
infringement, as Plaintiffs’ theory would allow, would “create[e] a cause of
action for, in effect, plagiarism,” and would be incompatible with the careful
ways that courts have circumscribed the Lanham Act to avoid unduly burdening
competition and free expression. [Citing Dastar.]
Lookalike cases involving ads involve “appropriating the
identity and the goodwill of the famous plaintiffs—that is, pretending to be
Woody Allen—rather than merely engaging in the same trade while happening to
look like the famous plaintiffs.” And, the court noted, Allen’s cases involved
use of identity in ads, not his lookalike’s presence as part of a good or
service. “To allow Plaintiffs to protect the downstream uses of their voices
merely because Plaintiffs originated them would disrupt the ‘carefully crafted
bargain’ struck by patent and copyright law and ‘misuse’ the Lanham Act to ‘to
reward [artisans] for their innovating in creating a particular [work or]
device.’”
What we’re seeing, post-JDI, is the further
development of a “use as a mark” doctrine to cabin trademark’s nearly unchecked
expansion, unfortunately mostly without discussion of the relevant
considerations and how they relate to the purposes of trademark law. That’s
present here, yay! It would be great for more courts to admit that this not an
entirely empirical assessment, or at least it’s only empirical at the
categorical level. (See also the pre-JDI Louboutin v. YSL case,
the standout entry in the “ipse dixit use as a mark” cases.) In addition, JDI’s
language about how even partial TM use is bad is, as has always been apparent,
complete nonsense (in the absence of a materiality requirement).
False advertising: Two theories here. First, plaintiffs
alleged that marketing their voices under the names “Kyle Snow” and “Sally
Coleman” was literally false. The court disagreed: “Plaintiffs point to
numerous examples of Lovo marketing the voices as what they truly are—synthetic
‘clones’ of real actors’ voices.” Also, Lovo falsely allegedly stated that the
cloned voices “came with all commercial rights.” But even if that’s true, “such
misrepresentations do not concern ‘the nature, characteristics, qualities, or
geographic origin’ of Plaintiffs’ cloned voices.” (Dastar.)
Second, plaintiffs alleged that Lovo “confus[ed] potential
customers … as to [Plaintiffs’] affiliation with Lovo and the ability to use
the Lovo service in place of traditional access to these actors,” and that Lovo
“misrepresent[ed] that [Plaintiffs] have partnered with Lovo.” Those were not
actionable as false advertising; they were repeats of the false affiliation
claim, and “Plaintiffs cannot avoid the requirement of a protectable mark by ‘disguis[ing]
a § 43(a)(1)(A) trademark infringement and unfair competition claim as a false
advertising § 43(a)(1)(B) claim.’”
In addition, plaintiffs failed to identify any false
implication or consumer confusion from Lovo’s advertising, “as opposed
to being confused by the similarity of the synthetic voices to Plaintiffs’
voices when heard in the wild. In fact, Plaintiffs actually allege that Lovo’s
advertising was clear on the lack of connection between Plaintiffs and Lovo’s
voice clones.”
Even if Lovo made misrepresentations, plaintiffs didn’t
plead actual injury. Where “the defendant and plaintiffs are competitors in the
same market,” materiality and injury usually blend together, but plaintiffs
didn’t allege that their losses were related to actionable false claims in
Lovo’s advertising. Instead, their losses were allegedly due to the existence
of Lovo’s product: “Put differently, Lovo’s services are more desirable to some
customers because they are cheaper and more accessible. While Plaintiffs might
have a different cause of action for such competitive harm, it does not sound
in Lanham Act false advertising.”
Copyright: Four claims, only one of which goes forward. The
court approved of the sequence (1) complaint filed without any copyright
claims, (2) registration secured, (3) complaint amended to add copyright claims
as consistent with Fourth Estate, and I think I agree, since (1) was far
from a sham complaint. The discovery rule applied for statute of limitations
purposes,
Contractual defenses couldn’t be resolved on a motion to
dismiss, even without taking into consideration the chat messages exchanged by
the parties. Turns out, Fiverr’s TOS provide special rules for “Voice Over
Gigs” “that are inconsistent with a transfer of unrestricted copyrights.”
Despite a general work for hire clause, the TOS stated that buyers of
voice-over recordings are only “purchasing basic rights, … allowing them to
use the work forever and for any purpose except for commercials, radio, television
and internet commercial spots.” Buyers need to purchase higher-tier packages of
either “Commercial Rights,” in order to “promote a product and/or service,” or
“Full Broadcast Rights,” in order to use the recordings “in radio, television
and internet commercials.” Lovo didn’t do that.
First, Sage successfully pled a direct infringement claim
based on Lovo’s use of her actual voice recording in an investor presentation,
pitches to investors, at a conference, and in Lovo’s external marketing
materials. (Although this is for later, the fair use defense for comparative
advertising is a distinct subset of copyright fair use cases where defendants
generally win. How else are they supposed to substantiate their claim of being
as good as a human?)
Both plaintiffs alleged direct infringement based on Lovo’s
use of their recordings to train the Genny AI model. These claims failed “for
lack of adequate explanation in the complaint.” They didn’t explain “what
training is or how it works, even at a very high level of generality. The Court
therefore cannot ascertain or reasonably infer which exclusive rights Lovo
allegedly infringed, or how.” But amendment was possible, including using
information-and-belief pleading. (On the other hand, it might be hard to plead
what was outside the scope of the license, even as modified by the chats.)
Copyright infringement by the AI outputs: Here there was
really no hope. Sound recording copyright only covers exact duplication of the
fixed sounds, which was not alleged here as even a possibility for Genny. And
we know that a voice isn’t copyrightable, since that’s what gives them their
valid ROP claims.
Contributory infringement: based on AI outputs, therefore
failed.
NY ROP: Although NY’s ROP is narrower than other states’,
and only covers uses in, essentially, advertising, there was enough here to
proceed. (Query how much of plaintiffs’ damage comes from the advertising and
not the underlying product; also, later on the incidental use
defense—advertising for a product that does not itself violate the ROP is
protected where the advertising actually relates to the product—may come into
play.)
The statute of limitations is one year, but the
republication exception can restart the limitations period, and applied here.
But are digital replicas even covered? Lovo argued that NY’s 2021 statutory
protection for digital replicas of deceased persons meant that expressio unius applied
to exclude that from the main ROP. But that reasoning ignored context. “Just
before the amendment was proposed, New York courts held that digital replicas
of living persons—at least ones with a visual component—were already covered by
the law. There is no indication in the text of Sections 50 and 50-f, or in the
legislative history, that the amendment was intended to overturn this
precedent, nor to exclude the possibility of a similar holding with respect to
audio-only voice clones.” The same could be true of voice. “If anything, the
Court views ‘voice’ as having a broader scope than a term like ‘picture,’
because it cannot plausibly be read to refer to any particular form of media or
representative device. While Section 51’s statutory ROP “is to be narrowly
read” in light of its legislative history, it is nonetheless “not to be obeyed
grudgingly by construing it narrowly and treating it as though it did not exist
for any purpose other than that embraced within the strict construction of its
words.”
As to the substance, plaintiffs sufficiently alleged
recognizability and Lovo itself allegedly represented that its creations are
“practically indistinguishable from the ‘real’ voice.” While “New York courts
have consistently dismissed Section 51 claims based on the use of a fictitious
name, even if the depiction at issue evokes some characteristics of the person
or the person is identifiable by reference to external sources,” “those cases
all involve a fictional character sharing certain discrete attributes or traits
with a Section 51 plaintiff, not the use of the plaintiff’s portrait or voice.”
“Lovo cannot escape liability merely because it appended fictitious names to
those appropriated voices. To hold otherwise would carve out a massive,
judicially created loophole in the statute with no textual or doctrinal basis.”
Plaintiffs also adequately alleged use in both advertising
and trade. “Advertising purposes has been defined as use in, or as part of, an
advertisement or solicitation for patronage of a particular product or service,
and trade purposes involves use which would draw trade to the firm,” although
the statute doesn’t reach newsworthy uses or matters of public interest. Lovo didn’t
raise a First Amendment, newsworthiness, or public interest defense.
The court is a little wobbly on whether the underlying
product itself counts as “trade” purposes, though my reading of the NY cases is
that it can’t (emphasis added to last sentence):
Whether or not the solicitation of
investors itself counts as an “advertisement,” the function of the “investor
presentation, which was later posted publicly online, is plausibly understood
as promoting Lovo’s underlying product. The same goes for the use of Lehrman’s
voice in tutorials and promotional articles posted online. Moreover, even if
the voices were not used in formal advertisements or solicitations, they were
clearly used for commercial purposes, and to draw trade to the firm. It is
plausible to infer that, by illustrating the value of the product and helping
show prospective customers how to use it, Lovo used its publicly posted
tutorials to increase the appeal of its software, acquire subscribers, and
retain subscribers it already had. Plaintiffs allege even that Lehrman’s
cloned voice was Lovo’s default product and one of its self-described “best”
voices.
NY GBL: Partially survives, raising an interesting Dastar
preemption issue. Section 349 prohibits “[d]eceptive acts or practices in
the conduct of any business, trade or commerce or in the furnishing of any
service” and Section 350 prohibits “[f]alse advertising in the conduct of any
business, trade or commerce or in the furnishing of any service.” “To successfully assert a claim under either
section, ‘a plaintiff must allege that a defendant has engaged in (1)
consumer-oriented conduct that is (2) materially misleading and that (3)
plaintiff suffered injury as a result of the allegedly deceptive act or
practice.’ ”
Mostly, plaintiffs failed to identify material
misrepresentations to the public; they couldn’t rely on misrepresentations to
them because they were not acting as consumers in those transactions, and
Sections 349 and 350 do not reach such “narrow, private dispute[s].”
However, plaintiffs adequately alleged that Lovo materially misrepresented
the scope of the “commercial rights” that it promised to provide to its
subscribers, thereby making its offerings appear more attractive. Lovo’s
consumers thus “purchased … [but] did not receive a product with the full
value with unlimited usage rights, which would have been a product with
legitimately acquired and/or created voices.” Sections 349 and 350 aren’t
limited to “nature, characteristics, qualities, or geographic origin.”
Note: While statutory interpretation got us Dastar,
if it’s true that extending the Lanham Act to licensing claims would cause a
conflict with copyright law, it’s equally true that extending state law to
licensing claims would cause a conflict with copyright law, creating conflict
preemption. Cf. Jackson
v. Roberts (finding preemption of a ROP claim). But the court says that
plaintiffs “could not bring a similar GBL claim based on alleged
misrepresentations about a copyright license, as such a claim would be
preempted by the Copyright Act.” I am not sure about the difference between
advertising “commercial rights” or “unrestricted use” and advertising
“properly licensed,” but ok.
“Lovo promised its subscribers that they could use Lovo’s
voice clones without legal restrictions. While Lovo was correct with respect to
federal copyright and trademark law, it was incorrect with respect to New York
law. Lovo’s consumers could, like Lovo itself, be liable under Sections 50 and
51 of the NYCRL.” [Again, the language here is open, but I would argue that the
only liability for customers would have to be for their own uses in
advertising/trade.]
Lovo’s conduct with respect to its subscribers was “consumer
oriented” in that it was “directed to consumers” and had “a broader impact on
consumers at large.” Although the mere unauthorized use of the plaintiffs’
images would not be “ ‘consumer-oriented in the sense that it potentially
affects similarly situated consumers,” reselling the voices to third-party
consumers for downstream use by those consumers was relevantly different.
And plaintiffs adequately alleged that they suffered injury
from Lovo’s misrepresentations in the form of diverted customers and lost
sales. “While it is true that Lovo offered its voice clones at lower prices
than the services of traditional voice actors, Lovo was able to poach
Plaintiffs’ customers only because it purported to offer products that its
subscribers could legally use—that is, because it engaged in misrepresentations
made unlawful by Sections 349 and 350.”
Fraud claims failed because (1) plaintiffs didn’t adequately
plead damages; although they alleged customer diversion/harm to brand value,
they didn’t quantify the “true value of the recordings they sold to Lovo, as
opposed to what Lovo paid,” which is the measure of fraud damages in NY, and
(2) their fraud claims merely restated their breach of contract claims, which
isn’t ok in NY “when the only fraud alleged is that the defendant was not
sincere when it promised to perform under the contract.”
Unjust enrichment, conversion, and common law unfair
competition claims were preempted by the New York Civil Rights law, which
preempts “all common law claims based on unauthorized use of name, image, or
personality, including unjust enrichment claims.” Relatedly, unfair competition
is like the failed Lanham Act claims (except also requires bad faith), and
there were no facts alleged “supporting an inference that Lovo acted with the
intent of generating confusion or coopting Plaintiffs’ reputations (as opposed
to the desirable characteristics of their voices).”
from Blogger http://tushnet.blogspot.com/2025/07/ai-voice-cloning-opinion-narrows-claims.html
Rebecca Tushnet 