# bytedance/Lance

A 3B-active-parameter native unified multimodal model for image and video understanding, generation, and editing.

Repository: https://github.com/bytedance/Lance
Canonical: https://ross.abutalabs.com/products/bytedance-lance
Homepage: https://lance-project.github.io
Language: Python
License: Apache-2.0
License Family: permissive
Topics: image-editing, image-generation, image-understanding, video-generation, video-understanding, unified-multimodal-models
Last push: 2026-07-14T13:17:31+00:00

## Health v2 (maintenance only)
Score: 55/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 92, release rhythm 35, longevity 7
- inputs: {"age_days": 110, "days_push": 50, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1329, forks 95 (observed 2026-08-28T04:04:23.844651+00:00)

## What it is
Lance is a 3B-parameter native unified multimodal model from ByteDance for image and video understanding, generation, and editing, trained from scratch with a multi-task synergy recipe. It is released as a research artifact with inference and fine-tuning code, model checkpoints on Hugging Face, and vLLM-Omni support.

## Use cases
- generate videos from text prompts
- generate images from text descriptions
- edit images and videos with natural language instructions
- understand and reason about image and video content
- fine-tune a small unified multimodal model on custom data
- run multimodal generation on a limited GPU budget

## When to choose
- you need a single compact model handling image/video understanding, generation, and editing
- you want a research baseline for unified multimodal modeling under constrained compute
- you want to experiment with fine-tuning a 3B multimodal model

## When to avoid
- you need production-grade, polished output quality for commercial media generation
- you need high-resolution or high-framerate video (training capped at 768x768 images and 480p 12 FPS video)
- you only need a specialized single-task model like pure text-to-image

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing, video-processing, llm-inference
- domain: artificial-intelligence, machine-learning, computer-vision, image-processing
- platform: python
- tags: multimodal-model, unified-model, text-to-video, text-to-image, image-editing, video-editing, research-project, bytedance, 3b-parameters, video, gpu, linux

## Member repositories
- bytedance/Lance (main) score 55

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:23.844651+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 2026-08-30T04:45:38.894443+00:00, confidence not recorded.
  - readme: https://github.com/bytedance/Lance (fetched 2026-08-28T04:04:23.844651+00:00, sha 313f76d11796)
  - homepage: https://lance-project.github.io (fetched 2026-08-29T12:04:42.866847+00:00, sha 37466d4a72e0)
- Data as of 2026-08-30T08:39:29.467469+00:00.