Has anyone tried using Astra 2 to reconstruct small clipped / blown-out areas?
I’m not talking about recovering the original data — obviously, if a region is completely clipped at 255,255,255, that information is gone.
What I mean is generative reconstruction using the surrounding context.
For example, imagine a face where only a small part of the cheek is completely blown out, while the eyes, nose, jawline and the rest of the skin are still clearly visible. In theory, Astra 2 should have enough spatial and temporal context to infer a plausible continuation of the skin texture and facial geometry.
The same could apply to small clipped areas of a roof, wall, sky, clothing, etc.
Has anyone experimented with this in Astra 2? If so, did prompting help, and what kind of prompts worked best?
Something along the lines of:
“Reconstruct the clipped highlight on the cheek using the surrounding facial structure and skin texture. Preserve the person’s identity, facial proportions and lighting. Do not alter the rest of the image.”
Also, are there any plans for a dedicated (local) model or tool focused specifically on temporally consistent reconstruction of clipped highlights in archival video?
