ComfyUI Node
Generate Prompt
Generates prompts from your Prompt808 library. All settings are exposed as node inputs. Debug mode is in ComfyUI Settings > Prompt808.
Generate Prompt
- libraries
- prompt
- negative_prompt
- status
◄library(no libraries)►
◄prompt_typeAny►
◄archetypeAny►
◄archetype_influence70►
◄moodAny►
◄llm_modelNone►
◄enrichmentAny►
◄quantizationFP16►
◄temperature0.70►
◄max_tokens1024►
◄api_urlhttp://127.0.0.1:1234►
◄keep_model_loadedfalse►
◄prefix►
◄suffix►
◄seed0►
CategoryPrompt808
Inputs (16)
| Name | Type | Default | Description |
|---|---|---|---|
| librariesopt | P808_LIBRARIES | Connect a Library Select node for multi-library generation | |
| libraryopt | COMBO | (no libraries) | No libraries — open the Prompt808 sidebar (camera icon) to create one |
| prompt_typeopt | COMBO | Any | Prompt style (Cinematic, Documentary, etc.) |
| archetypeopt | COMBO | Any | Archetype to filter elements by |
| archetype_influenceopt | INT | 700–100 | How strongly the selected archetype biases element selection (0-100%). Controls two things: (1) per-category probability of picking an archetype-matched element vs. a random pool element, and (2) minimum inclusion rate for categories the archetype covers -- e.g. a pose-only archetype at 80% guarantees poses appear in at least 80% of prompts. At 100%, ONLY archetype-matched categories are included. Has no effect when archetype is None. |
| moodopt | COMBO | Any | Mood modifier for the generated prompt |
| llm_modelopt | COMBO | None | LLM model for prompt composition. API (recommended) = OpenAI-compatible server (LM Studio, Ollama, llama.cpp, vLLM) at the api_url below — typically 10-20x faster than the local HF path on the same GPU. None = simple mode (no LLM). The other entries are local HF models — useful for self-contained installs but slower than a llama.cpp/GGUF backend. |
| enrichmentopt | COMBO | Any | Creative enrichment level for LLM composition |
| quantizationopt | COMBO | FP16 | LLM quantization (FP16, FP8, 8-bit, 4-bit) |
| temperatureopt | FLOAT | 0.700.1–1.5 | LLM sampling temperature (higher = more creative) |
| max_tokensopt | INT | 1024128–2048 | Maximum tokens for LLM generation |
| api_urlopt | STRING | http://127.0.0.1:1234 | Server URL for API mode (used when llm_model is set to API). Works with LM Studio, Ollama, llama.cpp, vLLM, or any OpenAI-compatible endpoint. Default points to LM Studio's local server. Ignored when using a local HF model. |
| keep_model_loadedopt | BOOLEAN | false | Keep LLM offloaded to CPU RAM after generation (faster next run) |
| prefixopt | STRING | Text prepended to the generated prompt (e.g. LoRA trigger word) | |
| suffixopt | STRING | Text appended to the generated prompt (e.g. quality tags) | |
| seedopt | INT | 00–4294967295 | Random seed for deterministic generation |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| prompt | STRING | Generated prompt |
| negative_prompt | STRING | Negative prompt (terms to avoid) |
| status | STRING | Generation status and archetype used |