# Emu Series

Emu Series: Generative Multimodal Models from BAAI

Repository: https://github.com/baaivision/Emu
Canonical: https://ross.abutalabs.com/products/emu-series
Homepage: https://baaivision.github.io/emu2/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: foundation-models, in-context-learning, instruct-tuning, multimodal-pretraining, generative-pretraining-in-multimodality, multimodal-generalist
Last push: 2026-01-12T08:39:29+00:00
Link (homepage): https://baaivision.github.io/emu2/

## Health v2 (maintenance only)
Score: 57/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 62, release rhythm 35, longevity 82
- inputs: {"age_days": 1150, "days_push": 233, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1778, forks 83 (observed 2026-08-28T04:05:35.199148+00:00)

## What it is
Emu3 is a suite of state-of-the-art multimodal models from BAAI trained solely with next-token prediction, tokenizing images, text, and video into a unified discrete space. The repository provides model weights, inference code, and training/fine-tuning scripts for both generation (text-to-image, video prediction) and perception (vision-language understanding) tasks.

## Use cases
- generate images from text prompts without diffusion models
- generate or extend videos by predicting next tokens
- build a vision-language chatbot that understands images
- caption images with a multimodal model
- fine-tune a unified multimodal model on custom instruction data
- run multimodal inference with vLLM
- research autoregressive multimodal architectures

## When to choose
- you want a single autoregressive model for both multimodal understanding and generation
- you need open weights for text-to-image or video generation research
- you want to avoid diffusion or CLIP-dependent architectures
- you need fine-tuning scripts for multimodal instruction tuning

## When to avoid
- you need a lightweight model for CPU-only or edge devices
- you only need a production text-only LLM
- you prefer mature diffusion-based image generation tooling
- you need a hosted API rather than self-hosted weights

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, llm-training, image-processing, video-processing, nlp, machine-learning
- domain: artificial-intelligence, large-language-models, computer-vision, image-processing, deep-learning
- platform: python
- tags: multimodal, next-token-prediction, autoregressive, text-to-image, text-to-video, vision-language-model, transformer, model-weights, baai, image-generation, video-generation, visual-understanding, video, natural-language-processing, gpu, linux

## Member repositories
- baaivision/Emu (main) score 57
- baaivision/Emu3 (backend) score 50

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:35.199148+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:30:42.871205+00:00, confidence not recorded.
  - readme: https://github.com/baaivision/Emu (fetched 2026-08-28T04:05:35.199148+00:00, sha 6f16f3bae827)
  - homepage: https://baaivision.github.io/emu2/ (fetched 2026-08-29T10:12:18.093810+00:00, sha ff39044f53bb)
- Data as of 2026-08-30T08:39:29.467469+00:00.