You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The model conversion pipelines are not directly part of Mochi Diffusion. They depend entirely on packages from Apple (coremltools, ml-stable-diffusion, python_coreml_stable_diffusion), Hugging Face (diffusers, transformers, scripts), and others (torch, etc). As these packages get updated, there are frequently bugs introduced in how they inter-operate. We are not able to fix bugs in these packages since we are not their author/maintainer.
Please use the Mochi Diffusion Discord server at https://discord.gg/x2kartzxGv for reports about model conversion issues, and for discussions about models that you converted yourself failing to run in Mochi Diffusion. At the Mochi Diffusion Discord server, you have a good chance of finding someone who has encountered the type of issue you are having, and who has a temporary work-around that they can share with you. There is even a dedicated channel there at https://discord.com/channels/1068185566782423092/1147197089298792548.
With several of these packages changing every few days right now, the best path at this particular time is to use pre-converted models that are known to work with Mochi Diffusion. There are more than 100 of these working base models at https://huggingface.co/coreml-community. All of the official ControlNet model types converted for use with Mochi Diffusion are available at https://huggingface.co/coreml-community/ControlNet-Models-For-Core-ML.
Finally, please note that Mochi Diffusion does not support SDXL models when run under macOS 13.x. Support is slowly being added under macOS 14, but you need to build Mochi from source to get these features as they are added. Once SDXL is feature-complete and bug free, there will be a formal release of a new version of Mochi Diffusion with SDXL support running under macOS 14. Apple has chosen not to enable SDXL under macOS 13.x, and that limitation carries over to Mochi Difusion.
Crash Log
No response
The text was updated successfully, but these errors were encountered:
I collected stuff from my miniconda environment for doing models conversions at this link, if you want to try the packages/scripts that have been working for me:
It seems, at least for a while,that a default new conversion environment with current packages is working in almost all cases. I have deleted the files from the repo noted above. You should be good to go on your own.
Processor
M1 (or later)
Memory
8GB
What happened?
The model conversion pipelines are not directly part of Mochi Diffusion. They depend entirely on packages from Apple (
coremltools
,ml-stable-diffusion
,python_coreml_stable_diffusion
), Hugging Face (diffusers
,transformers
, scripts), and others (torch
, etc). As these packages get updated, there are frequently bugs introduced in how they inter-operate. We are not able to fix bugs in these packages since we are not their author/maintainer.Please use the Mochi Diffusion Discord server at
https://discord.gg/x2kartzxGv
for reports about model conversion issues, and for discussions about models that you converted yourself failing to run in Mochi Diffusion. At the Mochi Diffusion Discord server, you have a good chance of finding someone who has encountered the type of issue you are having, and who has a temporary work-around that they can share with you. There is even a dedicated channel there athttps://discord.com/channels/1068185566782423092/1147197089298792548
.With several of these packages changing every few days right now, the best path at this particular time is to use pre-converted models that are known to work with Mochi Diffusion. There are more than 100 of these working base models at
https://huggingface.co/coreml-community
. All of the official ControlNet model types converted for use with Mochi Diffusion are available athttps://huggingface.co/coreml-community/ControlNet-Models-For-Core-ML
.Finally, please note that Mochi Diffusion does not support
SDXL
models when run undermacOS 13.x
. Support is slowly being added undermacOS 14
, but you need to build Mochi from source to get these features as they are added. OnceSDXL
is feature-complete and bug free, there will be a formal release of a new version of Mochi Diffusion withSDXL
support running undermacOS 14
. Apple has chosen not to enableSDXL
undermacOS 13.x
, and that limitation carries over to Mochi Difusion.Crash Log
No response
The text was updated successfully, but these errors were encountered: