ComfyUI Node
Concept Survival Check
A ComfyUI node in Eric Prompt Enhancers with 9 inputs and 4 outputs.
Concept Survival Check
- clip
- report
- all_survived
- missing
- carried_by_json
◄prompt—►
◄llm_instructions—►
◄threshold0.62►
◄embedderauto►
◄methodembedding►
◄backendlm_studio►
◄endpointhttp://localhost:1234/v1►
◄model►
CategoryEric Prompt Enhancers
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | The rewritten prompt to check. | |
| llm_instructions | STRING | The llm_instructions output of the Wildcard Processor or Semantic Variation node that made the picks. | |
| threshold | FLOAT | 0.620–1 | Embedding method: a concept survived when its best-matching phrase scores at least this. Measured with Qwen3-VL-Embedding-2B on things, materials and places: faithful paraphrase 0.69-0.74, same scene with the pick dropped 0.43-0.60, unrelated prompt 0.26-0.45. Other embedders sit elsewhere: read the report, then set this. |
| embedder | COMBO | auto | auto = connected clip, else Qwen3-VL-Embedding, else open_clip. lexical cannot judge meaning. |
| clipopt | CLIP | Optional. Used as the embedder when `embedder` is auto or graph_clip. | |
| methodopt | COMBO | embedding | embedding = fast, no LLM; reliable for things, materials and places, NOT for relations such as 'lit only from below' (an embedding hears the topic, not the relation). llm = ask the LLM backend whether each idea is stated; reads relations; needs a model loaded. both = the LLM decides, the embedding score is shown beside it. |
| backendopt | COMBO | lm_studio | For method llm / both. |
| endpointopt | STRING | http://localhost:1234/v1 | For method llm / both. |
| modelopt | STRING | For method llm / both. Blank = whatever model is loaded. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| report | STRING | — |
| all_survived | BOOLEAN | — |
| missing | STRING | — |
| carried_by_json | STRING | — |