ComfyUI Node
toobusy Load CLIP
Loads safetensors / .gguf text encoders and grows the model to fit added-token fine-tunes (no token loss). For an LLM prompt enhancer (-> Text Generate): the encoder must be a GENERATION-capable family. ComfyUI auto-detects the architecture from the file, and only Gemma-family encoders expose generate(). A Llama checkpoint (Dolphin etc.) LOADS but routes to an encode-only wrapper with no generate(), so Text Generate fails. Use a Gemma 3 (4B/12B/27B) encoder — e.g. an abliterated GGUF for uncensored prompt expansion. The `type` dropdown is ignored for LLM files (architecture wins).
toobusy Load CLIP
- clip
◄clip_name▾►
◄typelumina2►
◄fit_model_to_filetrue►
◄device▾►
Categorytoobusy/Make
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| clip_name | COMBO | Text encoder file (safetensors or .gguf). Both folders are merged here. For an LLM prompt enhancer (Text Generate), use a Gemma-family encoder — Llama/Dolphin loads but can't generate. | |
| type | COMBO | lumina2 | Text-encoder architecture (same list as the core CLIPLoader). Ignored for LLM files — ComfyUI auto-detects from the checkpoint. |
| fit_model_to_fileopt | BOOLEAN | true | Grow the model's embedding to the checkpoint's vocab so an added-token LLM (e.g. Dolphin: 128258 vs 128256) loads with every token kept. Off = behave like the normal loader. |
| deviceopt | COMBO | 2 options: default, cpu |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| clip | CLIP | — |