All Tools
E
OtherFreeOpen Source
EMU3
Next-token prediction for image, text, and video
Apache-2.0
ABOUT
Image generation usually needs a diffusion UNet while VLMs need CLIP plus an LLM. Emu3 trains one transformer on discrete image, text, and video tokens so the same next-token model can caption, generate stills, and extend clips without a task-specific architecture.
INTEGRATION GUIDE
1. Generate images from text with Emu3-Gen through Transformers
2. Chat about images with Emu3-Chat, including the vLLM backend
3. Continue a video clip by predicting the next vision tokens
4. Instruction-tune Emu3-Stage1 with the released SFT scripts
TAGS
pythonmultimodalgenerationvision-languagevideobaaitransformers