UniPic

SkyworkAIMIT更新放缓

开源的多图编辑扩散模型，支持同时处理多张输入图像进行编辑与生成，适合需要批量或组合式图像编辑的场景。

Open-source SOTA multi-image editing model

暂无版本877star44fork同步于 2026-11-07 07:31 UTC

多图编辑扩散模型图像生成多模态生成图像编辑

对比 GitHub 官网 aigc diffusion-models image image-editing image-generation

怎么上手

由 README 提炼，执行前请确认

原文

README 未给出安装或运行命令，需进入 UniPic-1/2/3 子目录查看各自 README

硬件要求按 README 说明

README 未说明显存、GPU 型号或 CPU 运行要求

下载：源码 zip·GitHub Releases（安装包）本站不镜像安装包

项目说明

由 README 自动翻译提炼

没人工点评？写下你的经验

Skywork 开源统一多模态模型系列，主打多图编辑与生成理解一体

开源的多图编辑扩散模型，支持同时处理多张输入图像进行编辑与生成，适合需要批量或组合式图像编辑的场景。

核心能力

UniPic-3 支持 1–6 张输入图做单图编辑与多图合成
用 CM + DMD 蒸馏实现 8 步推理、12.5 倍加速
UniPic-2 基于 SD3.5M-Kontext 与 MetaQuery 架构做扩散后训练
UniPic-1 用 1.5B 参数自回归统一建模理解与生成
同一仓库提供模型权重与官方实现

容易踩的坑

三个子模型分属不同子目录，需分别阅读对应 README 获取用法；模型权重需从 HuggingFace Collection 另行下载；README 未说明依赖版本与默认端口

内容由模型根据 README 生成，可能不完整或不准确，以官方文档为准。看原文 README

用户点评

登录用户的真实使用记录，可按版本和硬件筛选

共 0 条

还没有人点评。第一个说出真实体验的人最有价值。

README 原文

未翻译

在 GitHub 打开

📝 Overview

Welcome to the Skywork-UniPic repository!
This repository hosts the model weights and official implementations of unipic unified multimodal series, featuring three distinct modeling paradigms:

UniPic-3 (README) — 🔥 Open-source SOTA Multi-Image Editing Model. Unified framework for single-image editing & multi-image composition. Supports 1–6 input images with flexible resolutions. 8-steps inference with 12.5× speedup via CM + DMD distillation.
UniPic-2(README) — SD3.5M-Kontext and MetaQuery variants based on Efficient Architectures with Diffusion Post-Training, delivering state-of-the-art performance in text-to-image generation, fine-grained image editing, and multimodal reasoning.
UniPic-1(README) — 1.5B parameters, Unified Autoregressive Modeling for joint visual understanding and generation, enabling a single transformer to handle both perception and synthesis tasks.

🔥 Latest News

Date	Update
2026-01-09	Released UniPic-3 — 🔥 Open-source SOTA multi-image editing model. Support single & multi-image editing, 1–6 inputs, 8-step / 12.5× faster inference
2025-08-13	Released UniPic-2 — Unified Model Weights with Diffusion-based Post-Training
2025-07-30	Released UniPic-1 — Autoregressive unified modeling from scratch

✨ Key Features

🎨 Text-to-Image Generation — High-fidelity synthesis from natural language prompts.
🛠 Image Editing — Seamless inpainting, outpainting, and object manipulation.
🖼 Image Understanding — Robust perception capabilities for various visual tasks.
⚡ Efficient Architecture — Optimized for both accuracy and deployability.

📜 License

This project is licensed under the MIT License — see the LICENSE file for details.