AI in graphic design has made variation, mock-ups, retouching and first-draft assets fast and cheap, but it has not replaced the brief, the judgement about which option is right, the design system that keeps a brand consistent, or the typography. The data agrees: in Figma's 2025 survey of 2,500 product builders, 78% said AI made them more efficient but only 32% said they could rely on its output, and the US Copyright Office ruled in January 2025 that prompts alone do not make you the author of an image.
- 78% of Figma's 2025 respondents said AI made them more efficient; only 32% could rely on its output.
- 624 prompts were not enough: the US Copyright Office refused to register an AI-made prize winner in 2023.
- Design fixation: Midjourney users sketched fewer, more similar ideas in a CHI 2024 study.
- 20 February 2026: India's IT Rules amendment on labelling synthetic content took effect.
- Human-directed, AI-built: brief, judgement, systems and type stay with people.
My own practice runs on a simple split that I call human-directed, AI-built: I set the brief, the constraints and the final call, and machines do the repetitive production.
That split is not a slogan. It falls straight out of the research and the law, and this essay walks through both, with the 2025 and 2026 developments that matter to anyone designing in India.
In 2022, Jason M. Allen's image Théâtre D'opéra Spatial became the first AI-generated work to win the Colorado State Fair's annual fine art competition. On 21 September 2022 he applied to register its copyright. Asked how he made it, he said he had entered revisions and text prompts in Midjourney "at least 624 times", then cleaned it up in Photoshop and enlarged it with Gigapixel AI.
The US Copyright Office agreed his Photoshop edits were his own authorship, but asked him to disclaim the Midjourney and Gigapixel parts. He refused. On 5 September 2023 the office's Review Board affirmed the refusal: the work could not be registered as submitted. Hundreds of prompts had bought him a prize, not an author's claim to the picture.
US Copyright Office Review Board: decision on Théâtre D'opéra Spatial (5 Sep 2023)What do AI design tools actually do well?
AI design tools are best at producing many plausible options quickly:
- background extensions and generative fill;
- product mock-ups and retouching;
- layout variants and resized formats;
- rough concept images.
These are tasks where the cost used to be hours of production and the judgement was already made.
The 2025 and 2026 tools
The 2025 and 2026 releases moved in that direction. Adobe's Firefly AI Assistant, shown in April 2026, takes a plain-language request and chains about 100 tools across Photoshop, Illustrator, Premiere and Lightroom, returning native PSD and AI files that a designer can keep editing.
Figma's tools generate layouts and prototypes from prompts. Canva and Midjourney cover the fast-concept end. In every case the output is a starting file, not a finished decision.
That matches my experience on the marketing side. At Gnani.ai, AI workflows cut our production cycle time by about 60%. My view is that this is the right job for AI: cut repetitive production and raise what a small, competent team can do, while the strategy and the final creative call stay with people.
| Design task | Where AI helps now | Where a designer still decides |
|---|---|---|
| Brief and positioning | Summarising research, listing competitor patterns | What the brand should mean, to whom, and what it must not be |
| Concept exploration | Dozens of rough directions in minutes | Which direction is right, and why the others are wrong |
| Logo and identity | Mood boards, colour and shape variants | Drawing, optical correction, distinctiveness, legal clearance |
| Typography | Pairing suggestions, quick specimens | Hierarchy, spacing, script support (Devanagari, Bengali), licensing |
| Photo editing | Generative fill, object removal, upscaling | Whether the edit is honest; whether it needs disclosure |
| Production and resizing | Format adaptation, versioning, batch exports | Checking the system holds at every size |
| Developer handoff | Generating code and specs from frames | Tokens, components and the rules that keep a product consistent |
Are designers actually satisfied with AI tools?
Designers are less satisfied with AI than developers are. In Figma's 2025 AI report, 69% of designers were satisfied with AI tools against 82% of developers, and only 31% of designers used AI for core responsibilities such as asset generation, compared with 59% of developers who used it for core work such as writing code.
The gap makes sense once you look at what each group asks AI to do. Code has tests: it runs or it fails.
A design has no test suite; its quality is a judgement about fit, meaning and craft, which is exactly the part AI is weakest at.
The same survey found 52% of people building AI products said design matters more for AI products than for traditional ones, and 95% said it matters at least as much.
See the data as a table
| Designers (%) | Developers (%) | |
|---|---|---|
| Satisfied with AI tools | 69 | 82 |
| Say AI improves work quality | 54 | 68 |
| Use AI for core responsibilities | 31 | 59 |
Source: Figma: 2025 AI report, perspectives from designers and developers · Figma's chart caption gives 67% for developer quality; the article text gives 68%.

Does AI make designers more creative or less?
AI tends to make individual outputs better while making the field more alike, and in visual ideation it can anchor designers on the first image they see. Two controlled studies show both effects with numbers.
60 participants, 20 per group, had 20 minutes to sketch as many ideas as possible for a chatbot avatar after seeing an example with 14 salient features. One group had no help, one used Google Images and one used Midjourney V4.
Both inspiration tools increased fixation on the example, with the AI group showing the strongest effect (Bayes factor 124), and the AI group's sketches had lower variety and slightly lower originality than the unaided group. The limit: a small sample, one task, one 2023-era image model, and Bayesian estimates rather than significance tests.
Read the paper (free)293 writers were randomly given no AI help, one AI-generated story idea, or a choice of five, and wrote eight-sentence stories that 600 evaluators then rated. With five AI ideas, stories were judged 8.1% more novel and 9.0% more useful, and the least creative writers gained most (about 10.7% and 11.5%). But AI-assisted stories were noticeably more similar to each other, a social dilemma in which each person improves and the collective pool narrows. The limit: short written stories, not visual design, and online participants rather than professionals.
Read the paper (free)Put together, the lesson for a design studio is concrete:
- Diverge first. Do rough sketches and directions before you open an image generator, so the first picture on screen does not set the ceiling.
- Then multiply. Use AI to multiply and refine the directions you chose.
And if every brand in your category uses the same tools and prompts, sameness becomes a strategic risk, which is the same "blanding" problem covered in why logos look dated.
Can you copyright an AI-generated design?
In the United States, you cannot copyright an image that comes from prompts alone. The Copyright Office's January 2025 report on copyrightability, which drew on more than 10,000 public comments, says protection depends on human authorship: your own expressive input that remains visible, or your creative selection, arrangement or modification of the output.
For a designer that splits the work cleanly:
- A logo you drew, with AI used for exploration, is yours in the normal way.
- An AI illustration dropped in unchanged gives you rights in the layout, not the illustration.
That matters for logos especially, since a client will want to register and enforce the mark.
Where India stands
India has not settled the output question yet. DPIIT's committee released Part 1 of a working paper on generative AI and copyright on 8 December 2025, covering training only.
It proposed a "hybrid model": a mandatory blanket licence for training on lawfully accessed works, with government-set royalties due once a model is commercialised and collected through a proposed Copyright Royalties Collective for AI Training (CRCAT), while Nasscom dissented in favour of a text-and-data-mining exception with opt-outs. Authorship of AI outputs was left for Part 2.
| Question | United States (Copyright Office, 2025) | India (as of Nov 2026) |
|---|---|---|
| Are pure prompt outputs protected? | No; prompts alone are not enough | Not yet decided; DPIIT Part 2 pending |
| Is AI-assisted human work protected? | Yes, for the human-authored parts | General authorship rules apply |
| Training on copyrighted work | Part 3 pre-publication report, May 2025 | Hybrid licence proposed, Dec 2025 |
| Labels on synthetic images | No federal rule | Required on platforms under IT Rules amendment, Feb 2026 |
Do AI images need to be labelled in India?
Yes, on platforms. MeitY's amendment to the IT Rules, notified on 10 February 2026 and in force from 20 February, requires intermediaries to label "synthetically generated information", meaning audio, visual or audiovisual content created or altered by algorithms so that it appears real, and large platforms must verify users' declarations with technical checks.
Platforms must also embed provenance metadata that users cannot strip, and some takedown orders now have to be acted on within three hours instead of 36.
Content Credentials
The technical layer for this already exists. C2PA, a standards body whose steering committee includes Adobe, Google, Microsoft, Meta, OpenAI, Sony, the BBC and TikTok, publishes the Content Credentials standard, now at version 2.3. It attaches a signed record of how a file was made and edited, which tools like Photoshop and Firefly can write.
For designers, the practical upshot is to keep Content Credentials on when you export AI-assisted work, and to disclose generative edits in anything that looks like a real photograph.

Is it ethical to train on, or copy, other designers' work?
Copying another designer's work as your own was considered wrong long before AI, and the Graphic Design Stack Exchange thread on the question draws the line where most professionals do: borrowing one or two elements is normal, but taking a design and swapping colours, fonts and logos is close to plagiarism. The better advice in the thread is to work out what makes the original work, then build your own answer. Generative tools blur that line because a prompt can name a living artist's style.
Here is how I'd handle it in practice:
- No living names. Never prompt with a living designer's name for commercial work.
- Keep records. Log prompts, references and edits, which is also what you need to show human authorship.
- Redraw. Treat AI output as reference material that you redraw, not as a final asset.
What should a designer learn now?
Learn the parts AI is weakest at: briefs, critique, systems, typography and the craft of saying no to a plausible option. Those are the skills that turn a hundred generated variants into one right answer.
- Write the brief before the prompt. State the audience, the job of the design and what it must not look like. A generator cannot infer positioning.
- Sketch first, generate second. The fixation study suggests early AI images narrow your options.
- Build the system. Tokens, type scale, grid, colour roles and components are what keep hundreds of AI-assisted outputs on brand. The typographic grid guide is a starting point.
- Own the type. Image models still mangle letters, and they do worse with Indian scripts. Typography stays manual.
- Redraw what matters. Logos, icons and key illustrations should be redrawn as vectors by a person, for quality and for authorship. See the logo design process guide.
- Keep provenance. Turn on Content Credentials and keep prompt logs.
The same habits of critique and system-building are the core of the argument in self-taught graphic design in India. The tool-by-tool workflow is in the AI design workflow guide, and the wider business case is on my applied AI page.

Content volume is not proof of good design any more than it is proof of good marketing. AI makes volume free, so the scarce thing is the judgement that picks, cuts and connects. That judgement is the job now.
What's new
- Adobe's Firefly AI Assistant chains about 100 tools across Creative Cloud from one prompt VentureBeat, Apr 2026
- India notifies IT Rules amendment requiring labels on AI-generated content Business Today, Feb 2026
- DPIIT working paper proposes a hybrid licensing model for AI training Storyboard18, Dec 2025
- US Copyright Office says prompts alone do not create authorship US Copyright Office, Jan 2025
Questions people ask
Will AI replace graphic designers?
It is replacing production tasks such as resizing, retouching and first drafts. Figma's 2025 survey found only 32% of respondents could rely on AI output, and only 31% of designers used it for core work. The judgement and system-building parts of the job are growing in value.
Can I copyright a logo made with Midjourney?
In the US, not if it came from prompts alone. If you substantially redraw, modify or arrange it, the human-authored parts can be protected. India has not issued rules on AI output authorship yet, so redraw any mark you intend to register.
Which AI tools do designers use most?
Adobe Firefly inside Photoshop and Illustrator, Figma's AI features, Canva and Midjourney cover most professional use. The tool matters less than having a brief and a design system before you start.
Do I have to label AI images on Instagram in India?
Under the February 2026 IT Rules amendment, platforms must label synthetic audio and visual content and verify user declarations. Declare AI-generated or AI-altered realistic images when you post them.
Sources
- US Copyright Office Review Board: Second Request for Reconsideration, Théâtre D'opéra Spatial (5 Sep 2023; 624 prompts, Photoshop and Gigapixel)
- Figma: 2025 AI report, perspectives from designers and developers (2025)
- Wadinambiarachchi et al.: The Effects of Generative AI on Design Fixation and Divergent Thinking, CHI 2024 (2024)
- Doshi and Hauser: Generative AI enhances individual creativity but reduces the collective diversity of novel content, Science Advances (2024)
- US Copyright Office: Copyright and Artificial Intelligence, Parts 1–3 (2024–2025)
- US Copyright Office: Copyright Office releases Part 2 on copyrightability (2025)
- Storyboard18: DPIIT releases working paper on copyright and AI (2025)
- Legal 500 (Mondaq): Exploring the DPIIT's working paper on generative AI and copyright (2025)
- Business Today: Govt tightens rules on AI-generated content (2026)
- C2PA: Content Credentials specification and steering committee (2026)
- VentureBeat: Adobe's new Firefly AI Assistant (2026)
- Graphic Design Stack Exchange: Is it ethical to copy the work of other designers?
