v2.4.1

[mashb1t]

What's Changed

• fix: adjust clip skip default value from 1 to 2 by @mashb1t in fix: adjust clip skip default value from 1 to 2 #3011
• fix: add type check for undefined, use fallback when no translation for aspect ratios was given by @mashb1t in fix: add type check for undefined, use fallback when no translation for aspect ratios was given #3025
• feat: build docker image tagged "edge" on push to main branch by @mashb1t in feat: build docker image tagged "edge" on push to main branch #3026

Full Changelog: v2.4.0...v2.4.1

---

This discussion was created from the release v2.4.1.

[mashb1t]

For the people wondering why Docker images haven't received the update for clip skip: here it is!
Plus: if you want to live on the edge there is now the docker tag edge as mirror of main. Not to be confused with latest, which is the latest SemVer vesion. You can find the docker images here.

[scriptr78]

Waiting for promt translation as in fork. It`s excellent option.

[mashb1t]

@scriptr78 i'd have already included it if Google services were available in China. As they're not either a local LLM for translation has to be used or we have to implement a library supporting multiple translation providers.

[k8kiss]

Is a VAE now requiered? I have NONE selected and get this, But most models work fine wihtout VAE, but now I am forced to pick a VAE and all my generations are messed up.
File "G:\foocus_OG\Fooocus\ldm_patched\modules\sd.py", line 470, in load_checkpoint_guess_config
vae_sd = ldm_patched.modules.utils.load_torch_file(vae_filename_param)
File "G:\foocus_OG\Fooocus\ldm_patched\modules\utils.py", line 22, in load_torch_file
pl_sd = torch.load(ckpt, map_location=device, pickle_module=ldm_patched.modules.checkpoint_pickle)
File "G:\foocus_OG\Fooocus\modules\patch.py", line 481, in loader
raise ValueError(exp)
ValueError: [Errno 2] No such file or directory: 'G:\foocus_OG\Fooocus\models\vae\None'

[mashb1t]

What did you do that you have "None" even as an option in the UI? Default is the one fromthe selected base model, but you can change it to another one. No VAE isn't possible currently.

[k8kiss]

Thanks for the quick reply. That's interesting. So what's the default VAE? I have no idea why is NONE an option, I literally just updated.
Here is the full log when I choose NONE
VAE loaded: None
Traceback (most recent call last):
File "G:\foocus_OG\Fooocus\modules\patch.py", line 465, in loader
result = original_loader(*args, **kwargs)
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\serialization.py", line 986, in load
with _open_file_like(f, 'rb') as opened_file:
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\serialization.py", line 435, in _open_file_like
return _open_file(name_or_buffer, mode)
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\serialization.py", line 416, in init
super().init(open(name, mode))
FileNotFoundError: [Errno 2] No such file or directory: 'G:\foocus_OG\Fooocus\models\vae\None'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "G:\foocus_OG\Fooocus\modules\async_worker.py", line 979, in worker
handler(task)
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "G:\foocus_OG\Fooocus\modules\async_worker.py", line 467, in handler
pipeline.refresh_everything(refiner_model_name=refiner_model_name, base_model_name=base_model_name,
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "G:\foocus_OG\Fooocus\modules\default_pipeline.py", line 250, in refresh_everything
refresh_base_model(base_model_name, vae_name)
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "G:\foocus_OG\Fooocus\modules\default_pipeline.py", line 74, in refresh_base_model
model_base = core.load_model(filename, vae_filename)
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "G:\foocus_OG\python_embeded\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "G:\foocus_OG\Fooocus\modules\core.py", line 147, in load_model
unet, clip, vae, vae_filename, clip_vision = load_checkpoint_guess_config(ckpt_filename, embedding_directory=path_embeddings,
File "G:\foocus_OG\Fooocus\ldm_patched\modules\sd.py", line 470, in load_checkpoint_guess_config
vae_sd = ldm_patched.modules.utils.load_torch_file(vae_filename_param)
File "G:\foocus_OG\Fooocus\ldm_patched\modules\utils.py", line 22, in load_torch_file
pl_sd = torch.load(ckpt, map_location=device, pickle_module=ldm_patched.modules.checkpoint_pickle)
File "G:\foocus_OG\Fooocus\modules\patch.py", line 481, in loader
raise ValueError(exp)
ValueError: [Errno 2] No such file or directory: 'G:\foocus_OG\Fooocus\models\vae\None'

[k8kiss]

I had a back up and was able to revert back, updated again and now the "default vae" option was available, which I think is the way it is supposed to be. And that was working fine, then I selected a checkpoint that needs a VAE, so I chose the SDXL VAE, Generated a few images, but now the None option is back on VAE and "default vae" option is no longer there. If I choose NONE, then I get the above error.

**update - I am not sure if it is just me, but just refreshing the models causes the VAE to show the None option, and if selected then it doesn't work.... ugghhh time to go back to older version. I was excited about Clip Skip

[mashb1t]

ahhhh, got it, will check the default options after model refresh again.
Referenced in issue #3043

[mashb1t]

Fixed in #3043 and merged to main