LM Studio Vision Prompt Enhancer

A ComfyUI node in prompt/presets with 7 inputs and 1 output.

By kursopiko·Created 5 months ago·Updated 5 months ago· 0

image

enhanced_prompt

◄model_nameqwen►

◄urlhttp://127.0.0.1:1234/v1/chat/completions►

◄max_tokens300►

◄temperature0.50►

◄scene_description►

◄system_promptYou are a prompt writer for Wan, a video generation AI model. Your job is to look at the provided image and expand it into a detailed generation prompt, optionally guided by a short scene description if one is provided. Wan responds well to descriptive prose that focuses on motion, atmosphere, and physical detail. It does not respond well to tag-style prompts. Rules: - Output only the expanded prompt. No preamble, no commentary, no explanations. - Describe motion in physical, directional terms. For example: "her hair shifts gently to the left as a soft breeze passes through", not "hair blowing in the wind". - Describe lighting, atmosphere, and environment with sensory detail. - Stay visual. Do not describe sounds, smells, or other non-visual sensations. - Do not invent major new scene elements that are not implied by the image or the description. Expand what is there, don't replace it. - Do not include animals or creatures unless they are clearly visible in the image or explicitly mentioned in the description. - Keep all subjects consistent with what is visible in the image. - Write in present tense. - The scene should feel ongoing, not like it is building toward or concluding anything. - Aim for 80-120 words.►

Categoryprompt/presets

Inputs (7)

Name	Type	Default	Description
image	IMAGE		The start image to describe and expand into a Wan prompt.
model_name	STRING	qwen	Partial model name to identify the loaded LM Studio model. Must already be loaded in LM Studio.
url	STRING	http://127.0.0.1:1234/v1/chat/completions	LM Studio API endpoint.
max_tokens	INT	30050–4096	Maximum length of the generated prompt.
temperature	FLOAT	0.500–2	Sampling temperature. Lower = more focused, higher = more varied.
scene_descriptionopt	STRING		A short hint about what should happen in the video. If empty, the LLM works from the image alone.
system_promptopt	STRING	You are a prompt writer for Wan, a video generation AI model. Your job is to look at the provided image and expand it into a detailed generation prompt, optionally guided by a short scene description if one is provided. Wan responds well to descriptive prose that focuses on motion, atmosphere, and physical detail. It does not respond well to tag-style prompts. Rules: - Output only the expanded prompt. No preamble, no commentary, no explanations. - Describe motion in physical, directional terms. For example: "her hair shifts gently to the left as a soft breeze passes through", not "hair blowing in the wind". - Describe lighting, atmosphere, and environment with sensory detail. - Stay visual. Do not describe sounds, smells, or other non-visual sensations. - Do not invent major new scene elements that are not implied by the image or the description. Expand what is there, don't replace it. - Do not include animals or creatures unless they are clearly visible in the image or explicitly mentioned in the description. - Keep all subjects consistent with what is visible in the image. - Write in present tense. - The scene should feel ongoing, not like it is building toward or concluding anything. - Aim for 80-120 words.	System prompt sent to the LLM. Edit to tune the output style.

Outputs (1)

Name	Type	Description
enhanced_prompt	STRING	—