about apple/DFN5B-CLIP-ViT-H-14-384

[ffhelly]

Bro, I have a question.
I checked HF's website and did not find the 384 model, only the 378 model. Will this have an impact?
Also, where do I need to place this model in my directory? I am on Ubuntu system.

[hkchengrex]

The 378 model should be the same as the 384 model. I am not sure why two different numbers were used. If you go to https://huggingface.co/apple/DFN5B-CLIP-ViT-H-14-378, under "model usage", they also specify apple/DFN5B-CLIP-ViT-H-14-384.

Anyway, the model should be downloaded automatically by open_clip. Does it not?

[bbecausereasonss]

The install instructions don't mention most of these:

Prompt outputs failed validation
MMAudioFeatureUtilsLoader:

• Value not in list: vae_model: 'mmaudio_vae_44k_fp16.safetensors' not in ['mmaudio_large_44k_v2.pth']
• Value not in list: clip_model: 'apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors' not in ['mmaudio_large_44k_v2.pth']
• Value not in list: synchformer_model: 'mmaudio_synchformer_fp16.safetensors' not in ['mmaudio_large_44k_v2.pth']

[hkchengrex]

@bbecausereasonss You are probably referring to https://github.com/kijai/ComfyUI-MMAudio.
Not part of this repo.