4 ms·
Very cool. Would be interesting to train a model on images with alpha channels so outputs would be automatically masked and more easily composable. But maybe m
by cwkoss 3y ago
Very cool. Would be interesting to train a model on images with alpha channels so outputs would be automatically masked and more easily composable. But maybe masking is so good these days that would be futile?
When a user does img-2-img on a layer does it use the context from other visible layers in the generation?
- dheera 3y agoFor composing this approach works pretty well, maybe the author should consider making a UI for it https://multidiffusion.github.io/ https://multidiffusion.github.io/
- mottiden 3y agoThanks for posting. Really interesting
- Zetobal 3y agoSegmentation is solved... https://github.com/RockeyCoss/Prompt-Segment-Anything https://github.com/RockeyCoss/Prompt-Segment-Anything
- michaelt 3y agoSegment Anything is neat, but segmentation is far from solved. If the user generates a picture of a horse and rider to add onto another composition - they probably want to include the saddle.
- GaggiX 3y agoSAM is also conditioned on points, if it's ambiguous what you want to mask you can add a point on the saddle and the model will add it without a problem, segmentation is pretty much solved, I agree with the parent post.
- mdp2021 3y ago> Would be interesting to train a model on images with alpha channels Would be even more interesting to get an ANN middle system of ontology of the (finally) represented content in order to change the single items. An internal representation of qualified structured items in space as part of the chain. Prompt > accessible internal representation > render.