Pixels Without a Muse: Can India's IP Law Protect AI-Generated Creative Works?

India's intellectual property framework — principally the Copyright Act, 1957, the Trade Marks Act, 1999, and the Designs Act, 2000 — was architected for a world in which human ingenuity was the irreducible precondition for creative output. Generative AI dismantles that assumption. The result is a widening juridical vacuum in which substantial commercial investments in AI-generated brand assets, advertising content, fashion designs, avatars, and marketing materials are either unprotectable or protected only precariously. The Government of India's position, articulated in Parliament as recently as February 2024, that the existing regime is "well-equipped" to address these challenges, is critiqued here as constitutionally convenient but analytically untenable. Drawing on comparative jurisprudence from the United States, United Kingdom, and European Union, as well as India's own nascent but rapidly developing case law — including the ANI Media v. OpenAI proceedings before the Delhi High Court and the Bombay High Court's personality rights judgments — this article maps the gaps in Indian IP law and proposes a framework for legislative and judicial reform adequate to India's ambitions as an AI-powered economy.

I. Introduction: The Prompt-Engineer's Dilemma

A brand director at a fast-growing Indian startup types a detailed prompt into Midjourney: "minimalist logo combining the aesthetic of the Ajanta caves with a silicon chip, indigo and gold palette, clean vector style." The result is striking — original-looking, commercially valuable, and generated in under thirty seconds. The startup uses it as its brand identity. A competitor notices. It reverse-engineers a near-identical mark using the same tool. The startup's lawyers are asked: Do we own this? Can we stop them?

The honest answer — and the one this article provides — is: probably not, and certainly not easily.

This is not an edge case. India's creative economy is undergoing a generative AI revolution. Brands, advertising agencies, fashion houses, digital content creators, and technology startups are integrating tools like Midjourney, DALL-E, Stable Diffusion, Adobe Firefly, and Runway into their workflows at scale. Virtual influencers — AI-generated personas with millions of followers — are signing brand deals. Fashion designers are generating entire collections algorithmically. Marketing agencies are producing AI-composed jingles, AI-drafted copy, and AI-designed campaign visuals. The economic stakes are enormous; the legal infrastructure to protect these investments is, at best, uncertain.

II. The Architecture of Indian IP Law and Its AI-Shaped Gaps

A. Copyright and the Human Author Requirement

The Copyright Act, 1957 is the primary instrument of protection for creative expression in India. Section 13 confers copyright on original literary, dramatic, musical, and artistic works. The critical operative concept — originality — has been interpreted by Indian courts to require a minimum degree of skill, labour, and judgment, a standard increasingly inflected with a creativity threshold following Eastern Book Company v. D.B. Modak, where the Supreme Court held that copyright requires "a minimum degree of creativity," not mere labour or industry.

Section 2(d) of the Act defines "author" across different categories of works. For computer-generated works, Section 2(d)(vi) designates "the person who causes the work to be created" as the author. This provision — unremarkable when enacted, now desperately contested — is the principal hook on which proponents of AI copyright protection in India hang their argument: the user who provides the prompt causes the work to be created; therefore the user is the author; therefore the work is copyrightable.

The argument is ingenious but ultimately unsatisfying. First, Section 2(d)(vi) was drafted for relatively rudimentary computational tools — database generators, CAD programs — where a human operator's choices were granularly embedded in the output. The prompt-to-image pipeline of a modern model is categorically different: the transformation from natural language prompt to pixel-level output is mediated by billions of parameters and stochastic processes that no human directed or could meaningfully be said to have authored.

Second, the Act's definition of authorship assumes a human whose creative expression is being protected, not merely a human who set a process in motion. Copyright protects expressive choices, not ideas or processes. If the expressive choices are made by the model rather than the human, the doctrinal justification for copyright — incentivising human creative effort — dissolves. Third, Section 57's moral rights regime becomes conceptually incoherent applied to an AI-generated work: an AI has no personality, honour, or reputation to protect.

The result is a doctrinal trap: Section 2(d)(vi) may nominally accommodate AI-generated works, but the broader architecture of the Act resists it.

B. The Trademark Dimension

The Trade Marks Act, 1999 protects distinctive signs capable of distinguishing the goods or services of one enterprise from another. Unlike copyright, trademark protection does not depend on a human author, making it nominally more hospitable to AI-generated outputs — a logo generated by DALL-E is, in principle, registrable if it satisfies the distinctiveness requirements of Section 9.

Complications remain. AI-generated logos, trained on millions of existing marks, may recombine familiar visual elements and fail the distinctiveness threshold. More acutely, AI models may generate outputs confusingly similar to existing registered marks without the applicant's knowledge, exposing them to cancellation or infringement claims they had no practical means to anticipate. Trademark law also does not protect the generative process itself — a competitor using the same tool to produce a similar mark may do so with impunity absent independent distinctiveness through use.

C. The Designs Act, 2000

The Designs Act, 2000 protects the visual features of industrial articles — shape, configuration, pattern, ornamentation — applied by an industrial process, conferring a monopoly for ten years, extendable to fifteen. For AI-generated fashion designs and surface patterns, the DA offers a potentially more robust avenue than copyright, since it requires novelty rather than human authorship in the copyright sense.

Practical limitations remain significant: registration is article-specific, making broad protection expensive; and the DA creates a copyright-design overlap problem — once a design is registered and more than fifty articles are manufactured, any existing copyright in the design as an artistic work is extinguished.

D. The Sui Generis Gap

Indian IP law offers no sui generis protection for databases or AI-generated outputs, unlike the EU's Database Directive model. India's 161st Parliamentary Standing Committee Report on Commerce (2021) recommended a separate category of IP rights for AI-generated works. The Government's response — via the Union Minister of State for Commerce and Industry in a written reply to the Rajya Sabha on 9 February 2024 — was that the existing regime is "well-equipped," with no proposal to create a separate right. This position is critiqued in Part VI.

III. Category-by-Category Analysis

A. AI-Generated Artworks

The global consensus, crystallised in Thaler v. Perlmutter in the United States, is that purely AI-generated images are not copyrightable without meaningful human creative expression. The DC Circuit affirmed in March 2025 that the Copyright Act requires human authorship; the US Supreme Court declined to review the decision in March 2026.

The Indian position is doctrinally ambiguous but directionally similar. The critical question is what level of human contribution suffices: a five-word prompt is almost certainly insufficient, while an elaborate process of generation, selection, iterative refinement, and creative arrangement may well be, because the human's expressive choices become the dominant contribution. The practical implication: document the human creative process — version histories, prompt notes, and evidence of post-generation editing strengthen any copyright claim.

B. AI Avatars and Virtual Influencers

AI avatars raise at least three distinct questions. First, ownership: the human-authored elements (character design, backstory, visual style by human artists) are copyrightable; the purely AI-generated elements are not, or not clearly. The practical solution is to keep human authorship dominant and documented, relying on trademark protection for commercial exploitation of the persona.

Second, what happens when an avatar mimics a real person's likeness? Indian courts have moved quickly here. The Bombay High Court's July 2024 ruling in Arijit Singh v. Codible Ventures & Ors. held that a celebrity's voice, vocal style, mannerisms, name, and likeness are protectable personality rights, enforceable against AI tools that recreate them without consent. The Bombay High Court extended this in October 2025 in Suniel Shetty v. John Doe & Ors., granting injunctive relief against unauthorised AI deepfakes and synthetic voice reproduction, with a "John Doe" order extending protection into the Metaverse and emerging digital environments.

Third, wholly synthetic virtual influencers — modelled on no real person — fall into a further gap, with protection available only through trademark law, copyright in human-authored design elements, and contractual terms.

C. AI-Generated Fashion Designs

India's fashion industry is the world's sixth-largest, and AI is transforming design processes at every level. Copyright in fashion is notoriously weak in India — garments are typically articles of manufacture, not works of art, unless design elements are separable. The Designs Act is the more appropriate vehicle: it does not require human authorship in the copyright sense, only novelty and application to an industrial article. The critical strategic point is the copyright-design overlap rule — once a design is applied to more than fifty manufactured articles, copyright protection as an artistic work is lost, so AI-generated fashion brands must register under the DA before large-scale manufacture.

D. AI-Generated Brand Assets, Logos, and Marketing Materials

This is the category of most immediate concern to businesses. Copyright in AI-generated marketing materials depends on the degree of human creative contribution — an extensively curated and refined logo may attract copyright in its human-authored aspects; a first-result prompt download almost certainly will not. Trademark protection remains available regardless of how a mark was created, provided it is distinctive and clears prior rights — making trademark the most reliable vehicle, but only after a careful clearance search.

The strategic prescription: treat AI generation as the beginning, not the end, of the creative process; document human creative decisions; file trademark applications promptly; and conduct rigorous clearance searches.

IV. The Training Data Problem: Copyright Infringement Going Upstream

The IP challenges of generative AI begin at the input stage. Training large models requires ingesting vast quantities of copyrighted material without rights holders' consent — the issue at the heart of ANI Media Pvt Ltd v. OpenAI Inc & Anr. (CS(COMM) 1028/2024), filed before the Delhi High Court in November 2024, India's first generative AI copyright infringement suit. ANI alleges its copyrighted news content, including paywalled material, was used to train ChatGPT without authorisation.

The case raises questions of first impression: does storing copyrighted material for training infringe the reproduction right under Section 14 of the Copyright Act, 1957? Does generating responses constitute a separate, actionable act of infringement? Does the Section 52(1)(a) fair dealing exception apply to commercial AI training? Presided over by Justice Amit Bansal, the Delhi High Court held extensive final arguments leading up to March 2026, with major industry interventions from the Digital News Publishers Association and the Indian Music Industry. The court has reserved judgment, in what will likely be the most consequential Indian intellectual property ruling since Eastern Book Company v. D.B. Modak.

ANI contends that web scraping, tokenisation, and vectors stored in database indexes constitute copying regardless of whether the model later reproduces any individual work verbatim. Conversely, OpenAI argues its models extract abstract, non-expressive statistical patterns rather than the creative expression itself, satisfying the core ethos of fair dealing under Section 52, and that restricting training would impede constitutionally protected access to information. This defense attempts to draw a sharp legal boundary against traditional Indian copyright precedents like Syndicate of the Press of the University of Cambridge v. B.D. Bhandari. While B.D. Bhandari firmly held that a commercial publisher cannot reproduce textbook content to create competing expressive consumer works (like guidebooks), OpenAI argues that intermediate AI tokenisation serves an entirely non-expressive, transformative purpose. Whether the Delhi High Court accepts this mechanical distinction remains a pivotal, live question.

V. Comparative Perspectives

The United Kingdom's Copyright, Designs and Patents Act 1988, Section 9(3), expressly vests authorship of computer-generated works in "the person by whom the arrangements necessary for the creation of the work are undertaken" — a targeted statutory fix India conspicuously lacks. The United States maintains, through the US Copyright Office's January 2025 guidance and Thaler v. Perlmutter, that prompts alone do not confer sufficient human control for authorship, while AI-assisted works with dominant human creative choices remain copyrightable in their human-authored elements. The European Union's AI Act (2024) requires disclosure of training data summaries, alongside a text-and-data-mining exception with an opt-out right for rights holders under the 2019 Copyright Directive. Japan has taken the most permissive approach, permitting unrestricted commercial AI training under its amended copyright law — an approach this article argues India should not follow wholesale, given its large creative workforce and constitutional commitments to protecting intellectual labour.

VI. The "Sufficiency" Thesis: A Critical Evaluation

The Government's position rests on two pillars: that Section 2(d)(vi) of the Copyright Act, 1957, already accommodates AI authorship, and that India's international treaty memberships (Berne, TRIPS, WIPO) ensure adequate protection. Both are weaker than acknowledged. Section 2(d)(vi) does not resolve whether autonomously-produced work meets the originality threshold at all — no statutory interpretation can manufacture originality where no human creative contribution exists. And international treaty obligations require protecting works that meet domestic criteria; they do not expand the definition of a copyrightable work beyond human-authored expression.

More fundamentally, the sufficiency thesis misidentifies the problem: whether existing law provides adequate incentives for investment in AI-generated content, adequate protection for creators whose works train AI models, adequate clarity for courts and businesses, and adequate deterrence against AI misuse. On each dimension, the existing framework is demonstrably inadequate — a conclusion the Parliamentary Standing Committee reached and the Government dismissed without substantive engagement.

VII. Reform Proposals

1. Statutory amendment distinguishing "computer-assisted works" (human creativity dominant, AI as tool) from "autonomously generated works" (no meaningful human contribution), mirroring the UK's CDPA approach with originality safeguards.
2. A text and data mining framework, modelled on the EU approach, permitting non-commercial research use subject to an opt-out, with commercial training requiring a licence or statutory collective licensing.
3. Mandatory disclosure requirements for AI providers on categories of training content used.
4. Codification of personality rights developed through the Arijit Singh and Suniel Shetty line of cases, with statutory remedies for AI likeness replication, voice cloning, and deepfakes.
5. A centralised AI IP advisory body, potentially within the Copyright Office, to guide registrability and maintain a public register of AI-generated works.
6. Judicial specialisation in AI and generative technology within India's IP courts and tribunals.

VIII. Practical Guidance

Brands and marketing agencies: treat AI generation as the first step, not the final one. Build and document human creative decision-making into your workflow; file trademarks promptly; run clearance searches; confirm your AI vendor agreements establish output ownership and lawful training data provenance.

Fashion and design-led businesses: register AI-assisted designs under the Designs Act before commercial-scale exploitation, and stay alert to the copyright-design overlap rule.

Content creators and virtual influencer operators: document and register human-authored elements where possible; use trademark law for the persona's name and identity; obtain informed consent if modelling a real person's likeness.

Technology companies and AI developers: the ANI v. OpenAI proceedings signal judicial scrutiny of training data provenance is coming. Audit datasets and review Terms of Service for lawful sourcing.

IX. Conclusion: Pixels Without a Muse, But Not Without a Law

Generative AI did not ask India's permission before transforming its creative economy. The law, as always, is arriving after the fact. India's existing IP framework — adequate for the industrialised creativity of the twentieth century — does not satisfactorily address the conditions of the twenty-first. The "sufficiency" thesis offers false reassurance to businesses investing in AI-generated assets and false comfort to creators whose livelihoods are threatened by AI's ability to replicate their work at zero marginal cost.

The path forward is not despair but principled evolution. India's courts have already shown, through the personality rights jurisprudence of the Bombay and Delhi High Courts, a capacity to protect creative investment against AI exploitation by applying existing constitutional and tort principles with doctrinal intelligence. The legislature must now match this judicial agility. The ANI v. OpenAI judgment, when delivered, will be a watershed — and whatever it holds, it will expose the need for a comprehensive statutory framework governing AI and intellectual property. India has the opportunity to be not merely a consumer of global AI norms but an architect of them.

---

This article is intended for general informational purposes and does not constitute legal advice. Readers with specific queries are encouraged to seek professional legal counsel. © M&T Legal. All rights reserved.