画像
OmniGen
OmniGen: one unified model for text-to-image, editing, subject-driven and identity-preserving generation
★ 4.3k スターMIT
ガイドを見る →// image · ByteDance-Seed/Bagel
Bagel
Bagelは、ByteDance-Seedが公開しているオープンソースの画像プロジェクトです（GitHub: ByteDance-Seed/Bagel）。GitHubスターは6,191、フォークは544で、Apache-2.0ライセンスの下で公開されています。GitHubでの最終更新は2026年5月4日です。
ByteDance's open-source unified multimodal model for understanding, generation and image editing.
★ 6.2k スター⑂ 544 フォークApache-2.0更新 2026年5月
▦ ビジュアル — このプロジェクトのREADMEには埋め込み可能なスクリーンショットがないため、
ここではプレースホルダーを表示しています。下のガイド自体は完全な内容です。
ここではプレースホルダーを表示しています。下のガイド自体は完全な内容です。
// プロジェクトの基本情報
プロジェクト基本情報
- リポジトリ
- ByteDance-Seed/Bagel
- GitHubスター
- 6,191
- フォーク
- 544
- ライセンス
- Apache-2.0
- GitHubでの最終更新
- 2026年5月4日
- カテゴリ
- 画像
- 掲載タイプ
- 完全なインストールガイド
// 仕組み
仕組み
- Clarify your multimodal goal
Decide whether you need image understanding, text-to-image generation, image editing — or all three from one model.
- Clone and create the environment
Clone the repo and create the documented Conda environment with Python 3.10.
- Install dependencies and weights
Install the requirements and flash-attention, and download the BAGEL-7B-MoT weights from Hugging Face.
- Run the WebUI
Launch the Gradio interface and run your first understanding, generation or editing task.
// 機能
選ばれる理由
Unified multimodal model
One model covers visual understanding, text-to-image generation and image editing instead of separate specialist models.
Mixture-of-Transformers design
The 7B MoT architecture routes understanding and generation through specialized transformer experts.
Gradio WebUI included
A ready Gradio app (also hosted as a Hugging Face Space) lets you try the model interactively.
Training and evaluation scripts
Ships training guidance (TRAIN.md) and evaluation scripts for VLM, text-to-image and editing benchmarks.
Tunable generation controls
Documented CFG controls for text and image guidance, denoising intervals and timestep shift let you balance fidelity and prompt adherence.
Community deployment options
Dockerfiles and community ComfyUI and quantized (INT8/DF11-compressed) variants are linked from the README.
Advertisement
// 自分で実行
自分で実行してみる
- Clone the repository
Clone BAGEL (note the repository directory name) and enter it.
git clone https://github.com/ByteDance-Seed/Bagel.git cd Bagel - Set up the environment
Create the Conda environment and install dependencies including flash-attention.
conda create -n bagel python=3.10 -y conda activate bagel pip install -r requirements.txt pip install flash_attn==2.5.8 --no-build-isolation - Launch the WebUI
Download the BAGEL-7B-MoT weights from Hugging Face as the README directs, then start the Gradio WebUI to run inference.
python app.py # see README for the exact command
// 活用例
作れるもの
- Researching unified understanding-and-generation architectures
- Generating and then editing an image in one model session
- Benchmarking an open multimodal model against hosted VLMs
- Building an image-editing demo on the included Gradio app
Advertisement
// よくある質問
よくある質問
Who maintains BAGEL?
BAGEL is developed by ByteDance-Seed, ByteDance's Seed research team.
Is BAGEL free?
The code is Apache-2.0 licensed and the model weights are published on Hugging Face; check the model card for the weights' terms.
Does it need a paid API key?
No — you download the weights and run inference locally on your own GPU.
How does it compare to Qwen2.5-VL or Janus-Pro?
Those focus mainly on understanding (or separate generation); BAGEL is designed as a single unified model for understanding, generation and editing, and the README publishes its benchmark comparisons.
How do I get started?
Clone the repo, create the Conda env, install requirements and flash-attention, download the weights, and launch the Gradio WebUI.
// さらに探索
似ているツール
画像
HunyuanImage-3.0
Tencent's HunyuanImage-3.0 — a powerful native multimodal model for image generation.
★ 3.3k スターNOASSERTION
ガイドを見る →画像
pydantic-ai
How Python does AI. Agents, realtime voice, image generation, embeddings. Every model, every interface, typed end to end.
★ 20.4k スターMIT
ガイドを見る →画像
Janus
DeepSeek's Janus series - unified multimodal models that both understand and generate images
★ 17.8k スターMIT
ガイドを見る →画像
LLM-Hub
Run LLMs plus image, video and music generation locally on your phone
★ 603 スターCustom
ガイドを見る →画像
mmagic
MMagic — OpenMMLab's generative-AI toolbox: diffusion, text-to-image, restoration and a model zoo.
★ 7.5k スターApache-2.0
ガイドを見る →// 代替候補
代替候補： 画像
LibreAIカタログに掲載された画像プロジェクトの中で最もスターの多いもの — Bagelが合わなかった場合に検討したい、現実的な候補リストです。
- Stable Diffusion Webui★ 165.2k
Stable Diffusion web UI
- ComfyUI★ 136.0k
The node-graph powerhouse for precise, professional AI image and video generation.
- Immich★ 115.5k
High performance self-hosted photo and video management solution
- PhotoCraft★ 73.2k
PhotoCraft — a clean-room Photoshop reimplementation in pure Rust: layers, masks, adjustment layers, type, vectors, brushes and real PSD files, offline, MIT / Apache-2.0
- Fooocus★ 53.3k
Focus on prompting and generating
// さらに見る image
さらに見る： 画像
カタログの中でBagelとスター数が近い画像プロジェクト — 前後の棚もそのまま見て回れます。
- gimp★ 6.5k
Read-only mirror of https://gitlab.gnome.org/GNOME/gimp
- ChaiNNer★ 6.1k
A node-based image processing GUI aimed at making chaining image processing tasks easy and customizable. Born as an AI upscaling…
- photoview★ 6.5k
Photo gallery for self-hosted personal servers
- StableSwarmUI★ 4.9k
Modular Stable Diffusion web UI that makes power tools accessible, fast, and extensible
- aidea★ 6.9k
An APP that integrates mainstream large language models and image generation models, built with Flutter, with fully open-source code.
- DragGAN★ 4.9k
Unofficial DragGAN implementation: drag points on a GAN image to edit it interactively