Collected sources and patterns will appear here. Add from search or the patterns library.
(BoundingBoxes, TextEmbeddings) -> GroundingTokens
Encode spatial coordinates (e.g., bounding boxes) into Fourier features and merge them with corresponding semantic text embeddings to yield localized grounding tokens.
Problem it solves
Standard text embeddings lack explicit spatial coordinate priors, preventing precise layout conditioning in diffusion models.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.