
MMagic
β Text source-checked Nov 11, 2026Open-source AIGC toolbox for AI researchers and ML engineers to explore image and video processing, editing, and generation.
Source-checked against the vendor's site Β· rebuilt from the vendor's own pagesProfile updated Link checked How we rate
In short
MMagic is a free, open-source PyTorch toolbox for AI researchers to explore image and video generation, editing, and processing using various generative models.
Professional AI researchers and machine learning engineers
MMagic is an open-source AIGC toolbox designed for professional AI researchers and machine learning engineers to explore image and video processing, editing, and generation. The framework supports various fundamental generative models, including unconditional GANs, conditional GANs, internal learning, and diffusion models. Supported applications include text-to-image, image-to-image translation, 3D-aware generation, image and video super-resolution, video frame interpolation, inpainting, matting, restoration, colorization, and image generation.
MMagic is based on PyTorch and is part of the OpenMMLab project. Using the MMEngine and MMCV libraries from the OpenMMLab 2.0 framework, MMagic refactors data samples to support batch dimension manipulation and unifies data formats for training and inference. The framework supports the evaluation of generation-type metrics such as FID and reconstruction-type metrics such as SSIM, allowing for the evaluation of multiple datasets simultaneously.
The main branch of MMagic requires Python 3.9 or higher and PyTorch 2.0 or higher. Installation of MMagic depends on PyTorch, MMEngine, and MMCV.
The project is released under the Apache 2.0 license.
According to MMagic: MMagic is an open-source AIGC toolbox designed for professional AI researchers and machine learning engineers to explore image and video processing, Editing, and generation. Its pages list this strength: Supports popular applications including image restoration, text-to-image, 3D-aware generation, inpainting, matting, and super-resolution.
β The AI Tool Bible editorial team
Pros
- β The vendor says that MMagic provides state-of-the-art generative models for processing, editing, and synthesizing images and videos.
- β Supports popular applications including image restoration, text-to-image, 3D-aware generation, inpainting, matting, and super-resolution.
- β Decomposes the editing framework into modules that can be combined to construct customized editor frameworks.
- β Supports distributed training for dynamic architectures through MMSeparateDistributedDataParallel.
Facts
- Platforms
- cli, api (not source-checked)
- Pricing
- free - Free and open source (Apache 2.0) (not source-checked)
- API available
- Yes
- Open source
- Yes (license: Apache-2.0)
- GitHub stars
- 7,482 (checked Nov 13, 2026) (not source-checked)
- Last GitHub push
- Aug 6, 2024 (not source-checked)
- First commit
- Aug 2019 (not source-checked)
- Model used
- Multi-model (Stable Diffusion, ControlNet, StyleGAN, GANs, diffusion) (not source-checked)
- Vendor pages
- Changelog
Where this tool sits
Use cases
Frequently asked
- How much does MMagic cost?
- MMagic is completely free and open source. It is released under the Apache 2.0 license, allowing researchers and engineers to use it without any licensing fees.
- What technical requirements are needed to run MMagic?
- The main branch requires Python 3.9 or higher and PyTorch 2.0 or higher. Installation also depends on the MMEngine and MMCV libraries from the OpenMMLab 2.0 framework.
- What specific image and video tasks can MMagic perform?
- It supports text-to-image, image-to-image translation, 3D-aware generation, super-resolution, video frame interpolation, inpainting, matting, restoration, colorization, and general image generation.
- Which generative models does MMagic support?
- The framework supports unconditional GANs, conditional GANs, internal learning, and diffusion models. It is compatible with Stable Diffusion, ControlNet, StyleGAN, and other GANs.
- How does MMagic handle model evaluation?
- It supports evaluation of generation-type metrics like FID and reconstruction-type metrics like SSIM. The framework allows for the simultaneous evaluation of multiple datasets.