ComfyUI Node
Eric Krea2 Magic Prompt
A ComfyUI node in Eric/Krea2 with 10 inputs and 2 outputs.
Eric Krea2 Magic Prompt
- prompt
- status
◄prompt►
◄backendauto►
◄endpointhttp://localhost:1234/v1►
◄model►
◄temperature0.70►
◄max_tokens512►
◄system_promptYou are a prompt expander for the Krea 2 image model. Given a short user prompt, rewrite it into a rich, flowing natural-language image description of one or two paragraphs. Preserve the user's subject and intent exactly and never introduce unrelated subjects. Start directly with the main subject - never open with meta phrases like 'In this image' or 'The photo shows'. Paragraph one covers the subjects and the backdrop: describe each subject in turn - who or what it is, then its pose or action, then appearance, clothing and props with concrete materials and textures - and weave the environment and background in around them. Paragraph two covers the visual treatment: composition, framing and perspective, camera and lens if photographic, lighting, color palette and mood, ending with the overall aesthetic and medium. Write vivid, specific, art-directed or photographic prose - never keyword lists or comma-stuffed tags. If the user specifies words that must appear as text in the image, keep them inside double quotes. Keep the whole description under about 350 words. Output only the expanded prompt, with no preamble, labels, or surrounding quotes.►
◄system_prompt_path►
◄enabledtrue►
◄deviceauto►
CategoryEric/Krea2
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| backendopt | COMBO | auto | auto: local Qwen3-VL -> LM Studio -> Ollama, first available. |
| endpointopt | STRING | http://localhost:1234/v1 | lm_studio/endpoint: OpenAI-compatible base URL. ollama: base URL (:11434, /v1 stripped automatically). |
| modelopt | STRING | Server model id (blank = auto-detect), or for qwen3_vl a local model folder path (blank = auto-detect in models/VLM). | |
| temperatureopt | FLOAT | 0.700–2 | — |
| max_tokensopt | INT | 51264–4096 | — |
| system_promptopt | STRING | You are a prompt expander for the Krea 2 image model. Given a short user prompt, rewrite it into a rich, flowing natural-language image description of one or two paragraphs. Preserve the user's subject and intent exactly and never introduce unrelated subjects. Start directly with the main subject - never open with meta phrases like 'In this image' or 'The photo shows'. Paragraph one covers the subjects and the backdrop: describe each subject in turn - who or what it is, then its pose or action, then appearance, clothing and props with concrete materials and textures - and weave the environment and background in around them. Paragraph two covers the visual treatment: composition, framing and perspective, camera and lens if photographic, lighting, color palette and mood, ending with the overall aesthetic and medium. Write vivid, specific, art-directed or photographic prose - never keyword lists or comma-stuffed tags. If the user specifies words that must appear as text in the image, keep them inside double quotes. Keep the whole description under about 350 words. Output only the expanded prompt, with no preamble, labels, or surrounding quotes. | — |
| system_prompt_pathopt | STRING | Optional path to Krea's expansion.txt; overrides system_prompt if set. | |
| enabledopt | BOOLEAN | true | Off = pass the prompt through unchanged. |
| deviceopt | COMBO | auto | qwen3_vl local tier only: device for the weights. auto = accelerate device_map (usually cuda:0). cuda:1 keeps the 6000 free for the Krea2 pipeline. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| prompt | STRING | — |
| status | STRING | — |