WAN 2.2 Smooth Workflow v5.0 | AI Video Generator
MMAudio FeatureUtilsLoader
Already loads apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors
ComfyUI model file · CLIP vision encoder
Apple DFN5B CLIP ViT-H/14 (fp16) · Vision encoder for MMAudio video-to-audio
This is Apple’s DFN5B CLIP ViT-H/14 image encoder in fp16, packaged by Kijai for ComfyUI-MMAudio. MMAudio uses it to read your video frames before generating matching sound. Put it in ComfyUI/models/mmaudio and select it as clip_model in MMAudio FeatureUtilsLoader.
Kijai’s repo also has an fp32 copy (3.95 GB). fp16 is enough for MMAudio.
apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors
The short version
MMAudio generates audio that matches a video. To know what is on screen, it embeds the frames with a CLIP image encoder: Apple’s DFN5B ViT-H/14, a CLIP model trained on Apple’s filtered DFN-5B dataset. Kijai converted it to a single fp16 safetensors file for his ComfyUI-MMAudio nodes.
Searches often use the lowercase name apple_dfn5b-clip-vit-h-14-384_fp16.safetensors. Kijai’s file uses capitals; on Linux, rename it to match your workflow exactly.
Install
ComfyUI/ └─ models/ ├─ … └─ mmaudio/ └─ apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors
$ cd ComfyUI/models/mmaudio$ wget -O apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors \$ https://huggingface.co/Kijai/MMAudio_safetensors/resolve/main/apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors$ sha256sum apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors# d8ad903a79adc2b29bacfeedabfa3d78b03d36d3b4dcb885c0649fe8c5d763cb
Safetensors files hold tensors only and cannot run code. Still compare the SHA256: a failed download often leaves an HTML error page saved under the model’s filename.
Use
MMAudio FeatureUtilsLoader bundles three helper models from models/mmaudio: the audio VAE, Synchformer and this CLIP encoder. Its output feeds MMAudio Sampler with your video frames.
MMAudio FeatureUtilsLoader comes from the ComfyUI-MMAudio custom node pack. Install it with ComfyUI Manager first.
audio●
Frame size
The CLIP encoder resizes frames to 384×384 and reads them at 8 fps.
Not a ComfyUI CLIP Vision model
It does not go in models/clip_vision and Load CLIP Vision will not list it.
Works with
apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors is one part of the stack. A workflow that reports a missing model usually needs these too. The BigVGAN vocoder downloads itself on first run.
Run it
We match workflows by the model files they load: this exact file first, then other versions of it, then workflows built for the same use, then other models from the same family.
2 workflows
These workflows already load apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors. Open one and run it as is.
MMAudio FeatureUtilsLoader
Already loads apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors
MMAudio FeatureUtilsLoader
Already loads apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors
Compare
| File | Relation | Size | Best for |
|---|---|---|---|
| apple_DFN5B-CLIP-ViT-H-14-384_fp16.safetensors | This page · fp16 | 1.97 GB | MMAudio in ComfyUI |
| apple_DFN5B-CLIP-ViT-H-14-384_fp32.safetensors | Same encoder · fp32 | 3.95 GB | fp32 pipelines |
Troubleshoot
Why: The workflow asks for this exact filename, but ComfyUI cannot find it in models/mmaudio. Usually the file is missing, sits in a different folder.
Fix: Move the file into ComfyUI/models/mmaudio/, rename it to match the workflow (or pick it again in the dropdown), then refresh the node list or restart ComfyUI.
Why: MMAudio FeatureUtilsLoader is not built into ComfyUI. It comes from the ComfyUI-MMAudio custom node pack.
Fix: Install ComfyUI-MMAudio with ComfyUI Manager (or git clone it into ComfyUI/custom_nodes), then restart ComfyUI.
Why: The download is incomplete, or the browser saved an HTML sign-in or error page under the model’s filename.
Fix: Check that the file is exactly 1,973,509,450 bytes and that its SHA256 matches the value above. Download it again from Hugging Face · Kijai/MMAudio_safetensors if not.
Rather skip the setup?
Run WAN 2.2 Smooth Workflow v5.0 in your browser on a cloud GPU. Nothing to download.
FAQ
mmaudio_large_44k_v2_fp16, mmaudio_vae_44k_fp16, mmaudio_synchformer_fp16 and this CLIP encoder, all in models/mmaudio. The BigVGAN vocoder downloads itself on first run.ComfyUI/models/mmaudio/. Then refresh the node list or restart ComfyUI so MMAudio FeatureUtilsLoader lists the file.Sources: Kijai/MMAudio_safetensors · ComfyUI-MMAudio · apple/DFN5B-CLIP-ViT-H-14-378
RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.