# Fine-Tune Stable Diffusion Models with DDPO via Hugging Face TRL

_Tool · published 2023-09-29_

Hugging Face's TRL library now supports fine-tuning Stable Diffusion models using DDPO (Diffusion Direct Preference Optimization), giving creators a structured path to align image generation outputs with specific creative goals. The approach adapts preference optimization techniques — more commonly associated with large language models — to diffusion-based image generation workflows.

For film and post-production teams, this opens up practical options for training customized image models tuned to a production's visual style, asset requirements, or brand guidelines, without needing to build training pipelines from scratch.

## Sources
- [Hugging Face Blog](https://huggingface.co/blog/trl-ddpo)
---
Canonical: https://genbuzz.news/posts/fine-tune-stable-diffusion-models-with-ddpo-via-hugging-face-trl