M Magic (multimodal Aadvanced, Genergizing, and Iintelligent Creation) is an open source AIGC toolbox for professional AI researchers and machine learning engineers to process, edit, and generate images and videos.

MMagic allows researchers and engineers to use state-of-the-art pre-trained models and easily train and develop new custom models. Currently MMagic supports various image and video generation/editing tasks.

Best practices for master branch code are based on Python 3.8+ and PyTorch 1.9+ .

MMagic supports a variety of underlying generative models, including:

• Unconditional Generative Adversarial Networks (GANs)

• Conditional Generative Adversarial Networks (GANs)

• internal learning

• Diffusion model

• There are many other generative models coming soon

MMagic supports a wide variety of applications, including:

main features

• SOTA algorithm

  
  

  MMagic provides state-of-the-art algorithms for processing, editing, and generating images and videos.

• Powerful and Popular Apps

  
  

  MMagic supports the application of popular tasks such as image inpainting, graphic generation, 3D generation, image inpainting, matting, super-resolution and generation. In particular, MMagic supports the fine-tuning of Stable Diffusion and many exciting diffusion applications, such as ControlNet animation generation. MMagic also supports GANs for interpolation, projection, editing and other popular applications. Please start your AIGC exploration journey now!

• efficient framework

  
  

  Through the MMEngine and MMCV of the OpenMMLab 2.0 framework, MMagic decomposes the editing framework into different components, and can easily build a custom editor model by combining different modules. We can define the training process like building “Lego”, providing rich components and strategies. In MMagic, you can use different APIs to fully control the training process.benefited from MMSeparateDistributedDataParalleldistributed training of dynamic model structures can be easily implemented.