Copyright of AI images (India & global)
You generated a stunning render. The uncomfortable question your client will eventually ask: who actually owns it? In India, the honest answer is not the one you'd expect.
The Copyright Office registered an AI as co-author. Then it sent a withdrawal notice.
In a now-famous Indian case, an artist named Ankit Sahni filed to register a painting called 'Suryast', listing his AI tool RAGHAV as a co-author. The Copyright Office accepted it. Then, having second thoughts, it issued a withdrawal notice. That whiplash is the clearest signal you'll get about where Indian law actually stands: under the Copyright Act, 1957, an author must be a human. A machine cannot hold the pen. So when your client falls in love with an AI courtyard render and asks 'is this ours, can we use it, can anyone copy it?', the grown-up answer starts here.
Authorship is human-only — and that changes everything downstream
India's Copyright Act, 1957 ties authorship to a natural person
The whole question turns on one word: author. The Copyright Act, 1957 is the governing law in India, and Section 2(d) ties authorship to a human being. There is no provision for a machine to be the author of a work.
Follow that to its conclusion. A render produced by typing a prompt into Midjourney, with no meaningful human creative input beyond the words, is an output with no clear human author — and therefore, very likely, no valid copyright owner in India. It isn't 'owned by the AI company', and it isn't automatically 'owned by you'. It may sit in an awkward zone where nobody holds enforceable copyright at all.
That is not a loophole to celebrate. It means a beautiful image you generated might be one that anyone could legally copy — including the competitor pitching against you next week.
The first question is never 'is it good?' It's 'whose creative judgement made it?'
Suryast, the withdrawal notice, and the 'skill and judgement' test
The RAGHAV / 'Suryast' case is the precedent every Indian designer should know by name. The Copyright Office first registered the AI app as a co-author alongside Ankit Sahni, then issued a withdrawal notice — in effect establishing that an AI cannot be an author in India. The human can be; the machine cannot.
So what makes you the author? The 'creative spark' test — also framed as 'skill and judgement'. Merely typing 'draw a modern villa' is not enough creative contribution to make you the author of the result. But where you bring substantial human creative input — heavy iterative prompting, deliberate composition, selecting and combining outputs, real editing and post-work in Photoshop, sketch input you drew — your claim to authorship over the final work gets much stronger.
Think of it as a spectrum, not a switch. One lazy prompt sits at the weak end. A render where AI was one step in a heavily human-directed process sits at the strong end.
DPIIT's Dec-2025 working paper means: don't treat today's answer as final
This is live ground. The DPIIT working paper on Generative AI and Copyright (December 2025) is the government openly grappling with exactly these questions — which tells you the law is actively evolving and the position could shift.
The practitioner's move is not to wait for certainty. It's to behave defensibly now: maximise your human creative contribution so any future regime is likely to recognise your authorship, document that contribution, and — separately from copyright — manage the commercial-use licence of each tool you use. Copyright (who owns it) and licence (what the tool's terms permit you to do with the output, and whether they'll defend you if sued) are two different questions, and you must clear both.
And because the picture changes, name tools dated. What's true of a 2026 licence may not hold in 2027 — re-check the terms each renewal.
Copyright asks 'who owns it?' The licence asks 'are you even allowed to sell it?' Clear both.
Your visualisations live inside a larger professional package, so the copyright on a single AI render rarely sinks a project — but it can embarrass you. If a developer markets a scheme using an AI render that anyone can lawfully copy, your 'signature image' has no exclusivity. Strengthen your claim by making AI one step in a human-directed pipeline: render your own Revit or SketchUp geometry, prompt and edit deliberately, and keep the layered source files. Then your authorship over the composited deliverable is defensible, and the AI pass is just a tool you held.
Interior work leans heavily on generated mood and styling imagery, so this hits you daily. If you hand a client a pure text-to-image board and they reuse it in their own marketing, you may have no copyright to enforce and they may have no clear right either. Two habits fix most of it: bring real creative judgement (your curation, your edits, your sourced product overlays) so authorship is yours, and use commercially-safe engines for anything client-facing. Set the expectation in writing that concept imagery is illustrative, not a licensed final asset.
As a solo or student you can't afford a lawsuit or a stolen signature image, so build the discipline early. Treat every client-facing AI image with two questions: 'did I bring real creative input?' (for authorship) and 'does this tool's licence let me use it commercially and will it indemnify me?' (for safety). When in doubt, use Adobe Firefly for the commercially-safe training data and the indemnity, and keep your prompts, edits and source files as evidence of your creative spark. Cheap insurance for an expensive risk.
Adobe Firefly
Commercially-safe engine + indemnity
Trained on commercially-safe data (licensed/Adobe Stock/public-domain) and Adobe offers IP indemnity for enterprise output. The safest default for client-facing commercial work. It won't manufacture authorship for you under Indian law, but it removes the training-data lawsuit risk that many rivals leave on your desk.
Midjourney / Stable Diffusion / FLUX licence terms
Read-the-terms checklist
Each engine's commercial-use terms differ and most do NOT indemnify you against third-party claims. Midjourney grants broad use to paid subscribers; open Stable Diffusion shifts responsibility to you. Re-read the licence at each renewal — terms date fast — and never assume 'I paid, so I'm covered'.
Provenance log (your own practice)
Authorship evidence
A simple per-image record: prompts used, source sketch/geometry, edits made, tool + version + date. This is what demonstrates your 'creative spark' if authorship is ever questioned, and what lets you disclose AI use honestly to a client. Not a tool you buy — a habit you keep.
Studio Matrx Design Ideas (recolour)
Human-directed edit, as a live example
A worked case of AI as one controlled step: it recolours a real room you supply, guided by your choices, rather than inventing an image from nothing. That human-in-the-loop framing is exactly the kind of directed contribution that strengthens an authorship claim.
“I made it with my paid Midjourney account, so I own the copyright on every image it generates — it's mine to license and defend.”
Paying for the tool buys you a usage licence, not authorship. In India, authorship requires a human's creative spark; a one-line prompt may not clear that bar, meaning the image could have no enforceable copyright owner at all. Worse, your right to use it commercially depends on the tool's own terms — and most do not indemnify you if a third party sues. Ownership and permission are separate questions, and a subscription settles neither cleanly.
Workshop — run the copyright + licence clearance on a real render
Take one AI image you'd actually hand a client and put it through the two-question clearance. You'll leave knowing whether you can defend authorship and whether you're even allowed to sell it.
One AI image you generated for a real or sample project, plus the terms page of whatever tool made it. Free.
AI IMAGE CLEARANCE CARD Image: ____________ Tool + version: ____________ Date: ____________ A. CREATIVE SPARK (authorship in India) [ ] Did I do more than type one prompt? prompts used: ____ [ ] Did I supply my own sketch / geometry / photo as input? [ ] Did I edit / composite / post-process the output? [ ] Could I describe my creative judgement to a judge? -> Spark: WEAK / MEDIUM / STRONG B. LICENCE (permission to use it) [ ] Tool terms allow COMMERCIAL use? yes / no / unclear [ ] Tool offers indemnity if sued? yes / no [ ] Any training-data / third-party risk flagged? -> Use: CLEAR / RISKY / STOP VERDICT: ship / strengthen first / replace
- 1Pick one image you'd genuinely put in front of a client, and fill in the header (tool, version, date).
- 2Work section A honestly. Tag your contribution WEAK, MEDIUM or STRONG — one throwaway prompt is WEAK, a heavily edited render off your own geometry is STRONG.
- 3Open the tool's actual terms page and work section B: does it permit commercial use, and does it indemnify you? Most don't indemnify — write down which.
- 4If spark is WEAK, strengthen it: redo the image with your own sketch/geometry as input and at least one real edit, then re-score.
- 5If licence is RISKY, swap engines — regenerate the client-facing version in a commercially-safe tool like Firefly.
- 6Write the verdict line and staple this card to the image in your project folder. That's your disclosure-and-defence record.
You’ll walk away with
A completed clearance card for a real deliverable — a one-page, repeatable test that tells you, every time, whether an AI image is yours to defend and yours to sell.
Two quick tests, if you have five minutes.
- 01Take one of your best pure-prompt images and ask yourself the creative-spark question out loud: could you honestly claim you authored it? Notice how a single prompt feels.
- 02Open the terms of the image tool you use most and search the page for 'commercial', 'indemnif' and 'liability'. Read what it actually promises — most promise far less than designers assume.
In India, copyright needs a human author — so a one-prompt AI image may have no enforceable owner at all. Strengthen your claim with real creative input and documentation, and clear the separate question of each tool's commercial licence and indemnity. The law is evolving (DPIIT, Dec 2025), so behave defensibly now and re-check terms each renewal.
Copyright Act 1957: author = human. RAGHAV/Suryast: the AI co-author registration was withdrawn — AI can't author in India. The creative-spark test: more human input = stronger claim. Licence is a separate question; Firefly indemnifies, many don't. The law is moving — document and re-check.
Who owns an AI-generated image in India?
Often, no one cleanly. The Copyright Act, 1957 requires a human author, so a purely AI-generated image with only a trivial prompt likely has no valid copyright owner. Where you contribute substantial creative input — heavy prompting, your own input image, real editing — you strengthen your claim to authorship over the final work. The RAGHAV/Suryast withdrawal confirmed an AI cannot itself be an author.
Can I use AI-generated renders commercially for client projects?
Usually yes, but it depends on the tool's licence, not on copyright. Check each engine's commercial-use terms and whether it indemnifies you if a third party sues. Adobe Firefly is trained on commercially-safe data and offers indemnity; many others grant use but offer no protection. Treat commercial permission and copyright ownership as two separate clearances.
What is the creative-spark test for AI art?
It's the 'skill and judgement' threshold for human authorship. Typing 'draw a cat' is too little creative contribution to make you the author. Substantial, deliberate human input — iterative prompting, composition choices, supplying your own sketch, meaningful editing — adds the creative spark that can make the resulting work yours under Indian copyright thinking.
_Ownership is one grown-up question; the next is sharper — what you must never feed a public model in the first place, and the Indian data law that now governs it._