✦
Summary
LLaDA-Image is a six-billion-parameter model family for image generation and instruction-guided editing. A single checkpoint can create an image from text or alter a reference while preserving relevant content, supporting realistic pictures, stylized designs, and Chinese or English text rendering.
Key Features
The same model supports text-to-image generation and reference-guided editing.
The project demonstrates Chinese and English text rendering.
LLaDA-Image is a 6B-parameter model family.
The recommended Base sampling configuration uses 50 steps.
LLaDA-Image-Turbo provides distilled generation and editing in a few steps.
The release describes Twin-DMD distillation for efficient sampling.
The model zoo lists BF16 and FP8 checkpoints.
Inference code and weights are released; training code is marked coming soon.
About LLaDA-Image
The family uses a unified diffusion framework and builds its visual prior through image-only pretraining before paired language supervision and joint generation-editing training. The Base model uses a longer sampling schedule, while the Turbo model applies Twin-DMD distillation for generation in a few steps.
LLaDA-Image is useful for developers building local generation and editing workflows with shared infrastructure. Public releases include Base and Turbo checkpoints, BF16 and FP8 variants, and Diffusers-based inference code. The repository identifies training code as forthcoming, so the available release should not be described as a complete training pipeline.