h1t
/

Text-to-Image
Diffusers
lora

Sampler and scheduler? CFG and the number of steps?

#1
by junguang - opened

After testing, it has little impact on the multiple artistic styles of XL models, thanks for your work with selfless sharing, recommended samplers and schedulers,?and generating step counts and CFG values?

junguang changed discussion title from Did this LORA have a strong impact on XL's style? to Did this LORA have a strong impact on XL multiple art styles??
Owner

We have not observed the situation you mentioned so far. Can you point out a specific example (e.g., a prompt that works well on SDXL but performs worse on TCD), so we can investigate the reasons behind it? One possible reason is the data we use to train TCD comes from LAION, so it may not include as many styles.

junguang changed discussion title from Did this LORA have a strong impact on XL multiple art styles?? to Sampler and scheduler? CFG and the number of steps?

We have not observed the situation you mentioned so far. Can you point out a specific example (e.g., a prompt that works well on SDXL but performs worse on TCD), so we can investigate the reasons behind it? One possible reason is the data we use to train TCD comes from LAION, so it may not include as many styles.

After testing, it has little impact on the multiple artistic styles of XL models, thanks for your work with selfless sharing, recommended samplers and schedulers,?and generating step counts and CFG values?

Sign up or log in to comment