Veil for platforms
Provenance for every picture you generate.
Veil is a layer that sits inside an image generation pipeline. Between steps it checks the forming image against the photos people have veiled, so a platform can credit the work it draws on and refuse to reproduce what it shouldn’t.
A layer, not a filter
Veil runs between denoising steps on the model’s own predicted image, so a reproduction is stopped mid-generation instead of being screened after the fact.
Provenance you can audit
Every checked step keeps its royalty sheet: which identities, which photos, which tiles, at what score. Exported as JSON alongside the heatmap, for whoever needs to see it.
Credit by contribution
Shares come from matched tiles, not a model’s say-so: each identity’s share of the evidence above τ, and its coverage of the picture. A basis for paying people whose photos shaped an output.
Your thresholds
τ, the block gate, the check cadence and the cost of every check are policy you set, with an estimate before each run.
On the way
Planned
Veil today attributes and blocks reproductions of veiled pixels. The rest of the protection people ask for is being built, and will be part of the platform agreement.
Likeness protection
Keep a generator from producing a person who has asked not to be produced, not only their photos.
Watermarking
A mark that travels with a veiled photo and with every output that credits it.
Detection
Find veiled photos, and outputs derived from them, out in the wild.
To talk about Veil for your platform, reach Rao Studios at rao.nyc.