BAGEL
ByteDance's end-to-end multimodal reasoning model for vision-language tasks
ByteDance's end-to-end multimodal reasoning model for image-text understanding and generation.
Open foundation models — the base layers everything builds on.
Key features
- Unified multimodal understanding and generation, 7B active of 14B total
- Text-to-image quality competitive with specialist generators
- Free-form image editing, multiview synthesis, world navigation
- Mixture-of-Transformer-Experts architecture; Gradio app and ComfyUI
Good for
- Image understanding plus generation in one model
- Free-form visual editing and manipulation
Stack & tags
Pythonmultimodalvlmtext-to-imageimage-editingopen-weightsunified-modelworld-model
Quick start
python app.pyFrom the project README on GitHub.