TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Ethics & Provenance · AGE 19—22 · FACTUAL · EVOLVING
Clear the rights before shipping
Clear model terms, reference permissions and client agreement before any AI-assisted image ships.
When to use
Every time an AI-assisted output leaves the studio: a client deliverable, a competition entry, a publication, a social post with the studio's name on it. Not when challenged. Before.
The method
At least three checks, recorded beside the file. Model terms: which model or service, what its terms say about commercial use of outputs, the date you read them. Reference permissions: every image you fed the model — whose is it, and do you have the right to use it that way? Your own site photographs, yes. A photographer's image from a magazine, no, not without permission. A client's existing drawings, only if the client agreed. Client agreement: does your contract or a written note cover AI use on this project, and the uploading of project material to third-party services? Three is the floor, not the list. Depending on the project, add: confidentiality and any NDA; personal data in the images; a real person's likeness; a trademark in frame; stock-image licences; who owns what an employee makes; where the provider stores what you uploaded. One line per check, per deliverable. If any line is blank, the deliverable waits. And if you are a student or a junior, "clearing" may not be yours to do: your job is to find the gap, write it down, and raise it with whoever signs.
Watch for this
The reference image is where clearance most often fails. Students clear the model's terms and forget that the photograph they dropped in for image-to-image belongs to someone. "I only used it as a reference" is not a right the law recognises in most places. The output may carry that image's composition, and its owner's claim, into your deliverable.
Try it
Run the clearance pass on one AI-assisted output from the last month. Then list every reference image in that workflow and mark the ones you had no right to feed the model. Most students find at least one. Then add the fourth check the project actually needed — an NDA, a face, a logo — and say who in the studio had the authority to clear it.
Prove it
Run an IP-clearance pass on one AI-assisted output — model terms, reference permissions, client consent, and whichever further checks the project calls for — point out which reference images in the workflow you had no right to feed the model, and say which gaps you could clear yourself and which you would have to escalate.
How it works
At least three bodies of rules meet here. The model's terms are a contract between you and the provider. The reference image is governed by copyright, which in India is the Copyright Act 1957, and whether feeding a protected image to a model is infringement has not been settled by an Indian court as of this edition — the Delhi High Court's 24 July 2026 ruling in ANI v OpenAI was interim and concerned text. The client relationship is governed by your agreement and by confidentiality — a separate doctrine from privacy, with its own card on this map, Client confidentiality is not privacy. Personal data, likeness and trademarks each bring a fourth body of rules when they are in the frame. This card is EVOLVING because most of these are moving. The habit is not moving. The Lab treats the clearance pass as a studio reflex, like keeping a drawing register, learned early.
What this idea builds on
What this idea opens up
- Nothing yet names this as a foundation.
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.