Shypd.aiThe largest open index of AI toolsCustom Diffusion is an open-source image generation tool that enables multi-concept customization of text-to-image diffusion models. It allows fine-tuning of models like Stable Diffusion with a few images, reducing storage and training time.
Trending
About
Custom Diffusion is an open-source tool developed by Adobe Research for multi-concept customization of text-to-image diffusion models, such as Stable Diffusion. It allows users to fine-tune these models with as few as 4-20 images of a new concept, significantly reducing training time to approximately 6 minutes on two A100 GPUs. The method is efficient, fine-tuning only key and value projection matrices in cross-attention layers, which limits the extra storage per concept to 75MB. Custom Diffusion supports combining multiple concepts, like new objects with new artistic styles, and includes a dataset of 101 concepts (CustomConcept101) along with SDXL integration. It provides scripts for single-concept and multi-concept fine-tuning, as well as optimization-based weight merging.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
