songweige/rich-text-to-image

Leverages cross-attention maps from diffusion models to spatially ground text tokens, then applies region-based diffusion with rich-text formatting attributes (font size, color, style, footnotes) to enable fine-grained control over token emphasis, precise color rendering, and localized artistic style application. Integrates with Stable Diffusion v1-5, SD-XL, and fine-tuned variants through HuggingFace, with native support for Automatic1111 WebUI and LoRA checkpoints.

801 stars. No commits in the last 6 months.