How to finetune qwen image edit 2509?
#3
by
Styldod
- opened
Great work.
What platform or open-source codebase did you use to fine-tune the LoRA?
I’m planning to fine-tune it for a different use case on my own dataset and want to start from the same setup.
Hi, I used ai-toolkit
The only thing I changed is the lr, I set it to 2e-4. The resolution is 1024. Depending on what you are training you may need more steps. I got away with 480 steps because the model was already somewhat capable of applying textures with a low success rate. Apply texture lora boosts that aspect of the model, it technically did not train it from scratch.