How ezioby/calligrapher works
Calligrapher is a diffusion-based text-image customization application. Users supply a source image, text, mask, and style reference; demos or batch scripts prepare these inputs, encode reference style, run masked FLUX inpainting, and return generated images. The documented modes include self-reference, cross-reference, non-text references, and multilingual generation. The graph reflects sampled entry points and model code; some utilities and the multilingual inference path are only partially sampled, so uncertain internal wiring is omitted.
Main entry points: calligrapher.py Style Adapter · gradio_demo_multilingual.py Multilingual Demo · utils.py Image Preparation · gradio_demo.py Basic Demo
Components 15
User Entry Points
- Basic Demo
gradio_demo.py - Mask Upload Demo
gradio_demo_upload_mask.py - Multilingual Demo
gradio_demo_multilingual.py - Self-reference CLI
infer_calligrapher_self_custom.py - Cross-reference CLI
infer_calligrapher_cross_custom.py
Input Preparation
- Image Preparation
utils.py - Glyph Rendering
utils_multilingual.py
Customization Model
- Style Adapter
models/calligrapher.py - Image Projections
models/projection_models.py - Style Attention
models/attention_processor.py
Image Generation
- Inpainting Pipeline
pipeline_calligrapher.py - FLUX Transformer
models/transformer_flux_inpainting.py
Other components
- User
- TextFLUX Weights
- Generated Images
Connections 24
- User → Basic Demo: provides inputs
- User → Mask Upload Demo: provides inputs
- User → Multilingual Demo: provides inputs
- Basic Demo → Image Preparation: prepares images
- Mask Upload Demo → Image Preparation: prepares images
- Multilingual Demo → Image Preparation: prepares images
- Self-reference CLI → Image Preparation: prepares images
- Cross-reference CLI → Image Preparation: prepares images
- Basic Demo → Style Adapter: generates
- Mask Upload Demo → Style Adapter: generates
- Multilingual Demo → Style Adapter: generates
- Self-reference CLI → Style Adapter: generates
- Cross-reference CLI → Style Adapter: generates
- Multilingual Demo → Glyph Rendering: renders glyphs
- Self-reference CLI → Generated Images: saves images
- Cross-reference CLI → Generated Images: saves images
- Basic Demo → Generated Images: returns gallery
- Mask Upload Demo → Generated Images: returns gallery
- Multilingual Demo → Generated Images: returns gallery
- Style Adapter → Image Projections: projects embeddings
- Style Adapter → Style Attention: installs adapter
- Style Adapter → Inpainting Pipeline: invokes
- Inpainting Pipeline → FLUX Transformer: denoises with
- Multilingual Demo → TextFLUX Weights: loads optional weights