(LoRA) Fine-Tuning FLUX.1-dev on Consumer Hardware
Datasets mentioned in this article 1
Collections mentioned in this article 1
More Articles from our Blog
Community
Hi! Great blog post, thank you for your work.
The link to the script train_dreambooth_lora_flux_nano.py(in this section) is not working - perhaps the branch on which it was developed is not merged yet ?
Article author
Typo: CLIP is not a text encoder :P This is good
Hi all, I managed to make it run on a dataset of numerical art of my own (mostly 512x512 flame fractals), and am glad to share the LoRA - happy to get some feedback on it ! I'll update the model card soon ...
Thank you team for the great work, easy to use !
·
Article author
Awesome! glad to hear it worked well with your flame fractal dataset, results in model card looks cool
good work!
·
I don't remember you being this cocky. This is how you lose respect.
Article author
The script was already available since October 2024, just made a more detailed version :P
i know, my research group provided it ;)
·
Article author
it's called Terminus Research, for what it's worth, but we contributed the original flux training script, which was then redone for inclusion by Linoy, with modifications. and then we ensured PEFT and Diffusers worked with quantised training on LoRA.
·
Article author
•
My bad. Great work.
✅ 1. Do these instructions work with Flux Schnell? ✅ 2. Can you export the merged model as a .safetensors file after merging a LoRA? ✅ 3. Does a LoRA merge increase the size-on-disk of the model? 4. I recently bought a refurbed RTX 3090 for AI workloads, will these instruction still work on the RTX 3090 or should I return it for a RTX 4090?
An LLM I asked stated yes for three but I wanted to ask the experts.
·
Article author
I would say it should work for Schnell but you might have to change the loss function a bit to account for the fact it's a timestep-distilled model.
Yes.
I don't think LoRA merge should increase the final state dict size.
The numbers shouldn't change much.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images