Different LoRas for different regions

[lordsoffallen]

Hey,

Thanks for the great work! I had a question regarding LoRas. Is it possible to use different LoRas for different regions? For instance, in the research paper we had an image of batman and superman. Assuming i had a two LoRas one for batman and one for superman, could they then be invoked during the regional denoising steps?

Thanks in advance!

[antonioo-c]

Hi, thanks for the interest. I will start looking into the multi-LoRA thing these days, and I will let you know if I have made any progress. Stay tuned!

[scarbain]

Hi, any update @antonioo-c ? :)

[antonioo-c]

@scarbain Hi, I have done some experiments, and I found that Regional-Prompting-FLUX does not work with multi-LoRAs. Once used together, the quality of each LoRA region degrades and does not perform as good as single-LoRA. It will take longer time to experiment this multi-LoRA feature.

But, Regional-Prompting-FLUX works perfectly fine with Multi-IPAdapters, it's working like a zero-shot multi-LoRA. I will put up a code for it in this week.

[LianShuaiLong]

@scarbain Hi, I have done some experiments, and I found that Regional-Prompting-FLUX does not work with multi-LoRAs. Once used together, the quality of each LoRA region degrades and does not perform as good as single-LoRA. It will take longer time to experiment this multi-LoRA feature.

But, Regional-Prompting-FLUX works perfectly fine with Multi-IPAdapters, it's working like a zero-shot multi-LoRA. I will put up a code for it in this week.

excellent work

[Decycle]

@scarbain Hi, I have done some experiments, and I found that Regional-Prompting-FLUX does not work with multi-LoRAs. Once used together, the quality of each LoRA region degrades and does not perform as good as single-LoRA. It will take longer time to experiment this multi-LoRA feature.

But, Regional-Prompting-FLUX works perfectly fine with Multi-IPAdapters, it's working like a zero-shot multi-LoRA. I will put up a code for it in this week.

Is it possible for you to share the code for the multi-lora inference? I have a research project on this topic and I want to try to get more results on Multi-Lora.