The Delhi High Court is hearing the ANI v. OpenAI litigation. This copyright dispute over unauthorised data scraping dominates attention as the first case of its kind in India, but a parallel problem has gone unnoticed in the fashion industry. Generative AI is pushing the Designs Act 2000 to its breaking point. This article argues that AI-generated designs fall into a legal vacuum: they cannot secure registration without a human author or physical article yet they may infringe already protected designs by training AI models on protected visual designs. Modern fashion houses rely on generative diffusion models and Generative Adversarial Networks (GANs) to rapidly produce hundreds of designs, textile motifs, and stitch patterns. This instant automation demonstrates the grey areas in the Designs Act, which was originally drafted for manual sketching and physical manufacturing.
The first barrier is of physicality. It is based on a joint reading of Section 2(a) and Section 2(d) of the Designs Act, 2000. Design protection is limited to only visual features under Section 2(d) and Section 2(a) requires that the visual features are used in an article of manufacture, applied by an industrial process. That is, for a design to be protectable the design must be incorporated in a physical product.
However, this legal framework works well for traditional physical fashion, but it is unable to handle new digital fashion designs and generative AI. Nowadays, AI models (such as Fashion Diffusion and Style3D AI) are creating digital clothing pieces and 3D garments as well as designing patterns for augmented reality and virtual avatars that will never be produced in reality.
Although the Calcutta High Court in NEC Corporation v. Controller of Patents and Designs has widened the scope of protection under section 2(d) to include Graphical User Interfaces (GUIs), the DPIIT has recently issued draft amendments to include virtual designs, but designs generated solely by AI still lack protection and are stuck in the legal vacuum. Thus, AI-generated designs are not protected under design law, yet they may simultaneously infringe on the rights of the original proprietors through training of generative modules.
A second challenge is related to the statutory originality and proprietorship under Sections 2(g) and 2(j) of the Designs Act. The Supreme Court in Eastern Book Company v. D.B. Modak[1]established the doctrine of "modicum of creativity" based on human mental judgment. Similarly, the Indian Copyright Office applied this rationale and revoked the co-authorship certificate for the application of “RAGHAV AI.” A descriptive text prompt is merely an instruction, not creative execution. Similarly, the findings established in Zarya of the Dawn and Thaler v. Perlmutter reasoned that users have no deterministic control over outputs generated. Since autonomous generation does not have a human author, the very concept of ownership disappears, and designs can become public and lack protection.
Third, Section 22 of the Designs Act, 2000, prohibits “fraudulent or obvious imitation," but this standard cannot be applied to imitation done through artificial intelligence. Case laws such as Whirlpool of India Ltd. v. Videocon Industries Ltd[2] dealt with infringement instances involving direct replication. However, generative AI works differently by taking a large scale of data and recombining its parts into new works instead of copying the exact work. The section was originally drafted for traditional imitation only; it cannot actually fit into the synthetic copying of designs done by AI. Moreover, statutory ceiling of ₹50,000 (₹25,000 per breach) is outdated for these large-scale modern digital fashion industries. These gaps explain why the DPIIT's proposal necessitates reevaluating terms like "design" and "article," in Sections 2(d) and 2(a), respectively, to include virtual designs and designs displayed on screens, thus removing dependence on a physical item. This proposal is also in line with international treaties such as WIPO Riyadh Design Law Treaty, 2024 and Hague Agreement.
Against this backdrop, the recent ruling in the NEC case, alongside the DPIIT's proposed amendments demonstrates judicial and administrative willingness to adapt the Designs Act. However, adaptation by courts and drafts alone will not settle these questions. Parliament should amend Sections 2(a) and 2(d) to cover virtual designs, and amend Sections 2(g) and 2(j) to state expressly whether, and to whom, ownership of AI-assisted designs vests. It should also revisit Section 22 so that penalties are proportionate and the position on training-data misuse is clear. The DPIIT should require disclosure of AI involvement at registration.
Nonetheless, mere disclosure doesn't clarify the owner of the AI-generated design or determine the level of AI involvement in the design. Section 9(3) of the UK's Copyright, Designs, and Patents Act, 1988 and Section 2(d)(vi) of India's Copyright Act, 1957, provide one possibility, determining the authorship of computer-generated work as belonging to “the person who causes the work to be created.” But this causation rule tells us at whose instance a design exists if no human is responsible for the design; it does not tell us when that turns out to be the case.The U.S. Copyright Office's approach to Zarya of the Dawn offers a more flexible alternative to any bright line percentage test. The graphic novel was protected, but each of the images created by Midjourney within the novel was not protected, as copyright does not cover the work of the AI system itself, but only the selection, arrangement, or modification of it by a human. In India, this "functionality test" would imply that if a designer simply instructs an AI model and takes over the generated design verbatim, they would not own the design right, but if the designer has used their own creativity to select, curate, combine or significantly modify the AI-generated motifs into a final "garment or pattern, then that arrangement would be protected as a design right and not the underlying AI-generated motifs.
Parliament does not have to opt for either of these two models; it instead should run them in series. First, the provisions of Sections 2(g) and 2(j) of the Designs Act should invoke the Zarya approach of “selecting and modifying” design, in line with the Modak “modicum of creativity” standard. At that point, it is crucial to ascertain whether the design has sufficient human creativity. Then a causation-based test like Section 9(3) of the UK's Copyright, Designs and Patents Act and Section 2(d)(vi) of the India's Copyright Act would apply to fix the designated proprietor as the party that directed the use of AI. This way, the problem of “how much AI is too much” is resolved through real creative assessment, rather than through presumption.
A similar system is needed under the Designs Act for tackling the training data challenge that is growing worldwide. The EU's Copyright in the Digital Single Market Directive (DSMD) allows for commercial text and data mining, subject to Article 4(3), which gives the possibility to opt out and reserve works. The EU AI Act does this by requiring the general-purpose AI developers to have a policy respecting these opt-outs (Article 53(1)(c)).
This mechanism can be directly incorporated into the provisions of Section 22 of the Designs Act, 2000 relating to the "piracy of registered designs". The current wording of Section 22 only imposes liability when a registered design is used to make a physical article that is sold as an "unauthorized use of the registered design" or "obvious imitation" of the registered design. But there is another way: mass scraping of protected design databases for training generative diffusion models goes to the physical application of the protected designs entirely, creating new and synthetic products that are a combination of the protected designs' aesthetic features without causing classical statutory infringement. But while Section 22's outdated statutory ceiling of ₹50,000 per design seems to be a small sum for physical copies, the same should be applied to algorithmic copies as well. The provision should require compliance with a registered proprietor's formal opt-out declaration, and require the disclosure by the AI developer of when registered design corpuses have been used to train the model.
Taken together, these gaps indicate a single fundamental failure: the Designs Act was created around a human hand and a physical item, whereas generative AI delivers neither. The ongoing litigation, ANI v. OpenAI, has made courts confront the application of the Copyright Act to generative AI and the ambit of Section 52 of the Copyright Act.
This is not a theoretical discussion; fashion houses have already started massively using AI-generated designs that are legally unregulated. This gap has to be addressed by legislative measures, such as amendments to Sections 2(a), 2(d), 2(g), 2(j), and 22 of the Designs Act, or else a growing range of designs generated by AI will fall beyond the scope of the Act's protective and enforcement measures.
Ultimately, the Delhi High Court's final verdict in ANI v. OpenAI will settle the copyright side of the debate. But the design side remains unresolved. Until Parliament acts, India's fashion industry will continue to face legal uncertainty.