In-Context-LoRA

Why we included this project

Text-to-image teams juggling storyboards, visual identity work, and virtual try-on often end up training a different model for each job. In-Context LoRA takes a different route: it packs condition images and target images into one composite and uses a text prompt to say what the task is, so a single fine-tuning setup covers all of those cases. The repository ships ten pretrained LoRA models, sample data, and ComfyUI workflows, and training is light enough to finish on one GPU in a few hours. Practical touches like a reference captioning prompt and a config for the AI-Toolkit trainer make it easy to reproduce results, so it's a good starting point for teams that want task-specific image generation without building everything from scratch.