Can you train for sdv15 ?
+1
+1
Hi there! We're currently training for SD1.5. We were hoping it would be done and ready for today but it's not behaving as well as the SD2.1 version is. :( The SD1.5 version only seems to match the faces about 25% of the time. We're trying to figure out why this is the case.
For the glory of AMD, we are waiting :)
Is it possible to also consider the Mac platform M1M2? When I try to start the webui on M2, it prompts that mediapipe==0.9.1.0 cannot be installed even though it is in the venv folder... It can be used once but ultimately causes Python to crash and requires a restart of the WEBUI. And Thank you very much for your contribution！
Hi! Any news about an SD 1.5 compatible model ? Would be awsome as the majority of models are 1.5 based
We've been checking the progress every day. Our latest is looking pretty good. I'm making safetensors for it in a little bit and handing them off to some teammates to do extra sanity checking.
We've pushed the checkpoints for the SD 1.5 version of Control MediaPipe Face!
We debated internally about holding off and letting the model train further, but decided ultimately that it is sufficiently useful as-is. We will continue to train the model for a while after this release to see if there are any gains to be had and, if there are, we'll release a v3. For now, please have fun with this and let us know about the neat things you make.
The converted model in ONNX feels great!