SDXL
Hello,
Do you have any plans for releasing a SDXL-based model?
Thanks
I do, but I'm facing a few issues, my current model is not super stable and sometimes doesn't work at all, I'm still looking for solutions.
Hi achiru! Is it coming out more like a canny model? Curious if we could get in touch and discuss! Thanks! :)
Right now I'm focusing on achieving good results with a full-size controlnet, when I'm satisfied with them I'm planning to try with a smaller network, but for now it's not perfect.
I'd be happy to chat and exchange ideas about it! (I have some social media linked to my profile to get in touch)
I have good news: I have a working model for SDXL, although I'm still not satisfied by the results. It works "fine" but is not to the level of SD1.5 V2 yet.
I'm still improving the model and hopefully I'll have a perfect version soon!
Should I release this version in the meantime?
SDXL training is so slow (except if you have tons of VRAM)...
@achiru Would be amazing to have an early version, similar to how Stability released their "SDXL 0.9" version, so folks can setup their pipelines to use it as you improve it towards "1.0"
I have good news: I have a working model for SDXL, although I'm still not satisfied by the results. It works "fine" but is not to the level of SD1.5 V2 yet.
I'm still improving the model and hopefully I'll have a perfect version soon!
Should I release this version in the meantime?
SDXL training is so slow (except if you have tons of VRAM)...
@achiru I'm assuming you're referring to the quality? because other SDXL controlnets that already exist have that exact flaw; the more they try to control SDXL to more far the quality is from how SDXL usually is. anyways it will be AMAZING for an early version to be released, it will help getting a better understanding for how to improve
Just wanted to check if there are any updates on this. No pressure, just curious about the current status.
V1 is out now, still missing the readme though
https://huggingface.co/monster-labs/control_v1p_sdxl_qrcode_monster