|
--- |
|
title: README |
|
emoji: 📉 |
|
colorFrom: purple |
|
colorTo: indigo |
|
sdk: static |
|
pinned: false |
|
--- |
|
|
|
FlowChef introduces a novel approach to controlled image generation by leveraging rectified flow models (RFMs) for **efficient, training-free, inversion-free, and gradient-free steering of denoising trajectories**. |
|
Unlike diffusion models, which demand extensive training and computational resources, FlowChef unifies tasks like **classifier guidance, inverse problems, and image editing** without extra training or backpropagation. |
|
By model steering facilitated by gradient skipping, FlowChef sets new benchmarks in performance, memory, and efficiency, achieving state-of-the-art results across diverse tasks. |
|
|
|
- [**Project page**] [https://flowchef.github.io/](https://flowchef.github.io/) |
|
- [**Paper**] [https://flowchef.github.io/static/docs/FlowChef_ArXiv.pdf](https://flowchef.github.io/static/docs/FlowChef_ArXiv.pdf) |
|
- [**Demos**] (find below) |
|
|
