Flame-Code-VLM

Why we included this project

Flame is a vision-language model built for a narrow job: reading a UI mockup and turning it into React code that is structured and reusable rather than a static approximation. The project pairs the model with the pipeline that produced it, including automated data synthesis that generates labeled training snippets from real code, a three-stage training process, and an evaluation suite that checks syntax, behavior, and visual fidelity. That makes it useful on two levels. Researchers building or fine-tuning image-to-code models can study the training approach and the released datasets, while front-end teams can experiment with design-to-code generation for React. Multi-image input is a practical detail, since it lets the model adjust as mockups change. It is a framework to study and adapt rather than a drop-in production tool, so expect to invest in setup and training before.