CLIP Vision missing or incompatible: it is not the prompt's text encoder

Symptoms and scope

Reference-image workflows need a vision encoder with the expected architecture. Distinguish it from the object feeding CLIPTextEncode.

Error fragments for search; IDs, values and filenames may vary:

ClipVision model not found.
clip_vision
CLIPVisionLoader

Source-supported context

IPAdapter Plus lists variant-specific vision encoders under models/clip_vision. Its pinned unified-loader source calls get_clipvision_file(preset), which searches the registered clip_vision filenames: a VIT-G preset takes the bigG branch, Kolors takes a separate branch, and other presets use a ViT-H-oriented pattern. If no candidate matches, the loader raises ClipVision model not found.. A matching filename alone cannot prove that the bytes contain the right architecture. Core CLIPVisionLoader is a separate node that reads a selected clip_vision file and can report an invalid vision model; a text-encoder CLIP does not replace it. IPAdapter README Pinned resolver Pinned unified loader ComfyUI core nodes

Cases to distinguish

1. The required vision file is absent from the running instance's clip_vision list, or another type of weights was placed there.

2. A filename matches the preset resolver but its content is invalid or its vision architecture does not match the adapter variant.

3. The author's recommended filename and the installed resolver's matching pattern differ from the local name; an existing file may not be discovered.

Diagnostic sequence

The sequence below is an editorial procedure based on the cited context, not a diagnosis already confirmed for your environment.

Step 1. Record the exact error and node class. Distinguish the unified loader's ClipVision model not found. from a core CLIPVisionLoader invalid-file error or a later embedding shape failure.

Step 2. Record the selected preset and the running server's clip_vision file list. Compare the installed resolver pattern, publisher source and actual file identity with the author's pairing instructions; a matching extension, size or renamed file is not enough.

Step 3. In a copy, test one reference image with unrelated style, FaceID and multi-image branches removed, using a publisher-supported vision encoder and adapter pair.

Step 4. If discovery succeeds but loading or dimensions fail, preserve that new error and inspect file integrity and architecture pairing. Repeated renaming cannot turn the wrong weights into a compatible encoder.

Completion check

The running instance discovers and loads the intended vision encoder, and a minimal image-guided workflow reaches sampling with the matching adapter. File discovery alone is insufficient.

Limitations and cautions

Restart or refresh the actual ComfyUI instance after changing model paths. The pinned resolver selects the first filename matching its pattern, so multiple candidates should be identified explicitly. Publisher links are not evidence that this site downloaded or verified weights.

Original sources

• ComfyUI IPAdapter Plus README — checked 2026-09-26.
• Pinned IPAdapter vision-file resolver — checked 2026-09-26.
• Pinned unified loader — checked 2026-09-26.
• ComfyUI core nodes — checked 2026-09-26.

Source and English-language review: 2026-09-26. No vision model was downloaded, loaded or run by this site; no GPU test or compatibility guarantee is claimed.

Related troubleshooting and guides

• IPAdapter model not found: unified presets and manual loaders use different rules
• mat1 and mat2 shapes cannot be multiplied: inspect the model component chain