ComfyUI Prompt Guide: Write Better Prompts for AI Image Generation

How Prompts Work in ComfyUI

Prompts are text descriptions that tell the AI what to generate. In ComfyUI, prompts are processed by the CLIP Text Encode node, which converts your text into numerical vectors the model understands.

There are two types:

• Positive prompt — describes what you want in the image
• Negative prompt — describes what you want to avoid

Basic Syntax Rules

Write in English

CLIP models are trained on English text. Write prompts in English for best results.

Prompt Structure

A well-structured prompt follows this pattern:

[Subject], [Features], [Environment/Background], [Style], [Quality modifiers]

Example:

1girl, long black hair, red dress, standing in a flower garden,
golden hour lighting, anime style, masterpiece, best quality,
highly detailed

Separate with Commas

Use commas to separate different concepts. This makes it easy to manage and adjust individual elements.

Comments

ComfyUI supports comments in prompts:

• Single-line: // this is a comment
• Multi-line: /* this is a block comment */

Useful for annotating complex prompts without affecting generation.

Weight Control

Syntax

Adjust how strongly the AI follows specific parts of your prompt:

Keyboard Shortcuts

Select a word or phrase in the text field, then:

Weight Guidelines

• Keep weights between 0.5 and 1.5 for best results
• Weights above 1.5 often cause visual artifacts
• Weights below 0.5 may have negligible effect

Random Selection

Use curly braces to make ComfyUI randomly pick one option per generation:

{red|blue|green} dress

Each generation will randomly choose one color. Useful for exploring variations quickly.

Prompt Position Matters

Keywords placed earlier in the prompt have slightly higher influence. Put your most important elements first:

portrait of a woman, detailed face, soft lighting, studio photo

Here, "portrait of a woman" has the strongest influence on the output.

Quality Enhancement Keywords

Add these to your positive prompt to improve output quality:

Common Negative Prompt Keywords

Add these to your negative prompt to avoid common issues:

low quality, worst quality, blurry, watermark, text, logo,
deformed, bad anatomy, extra fingers, extra limbs, disfigured,
cropped, out of frame

Model-Specific Prompt Tips

Useful Tools

Prompt Helper Plugins

• ComfyUI-Custom-Scripts — Autocomplete for embedding names in the text field
• ComfyUI-ArtGallery — Visual preview of style prompts
• comfyui-prompt-composer — Node-based prompt management

Prompt Reference Sites

• Civitai — Browse model pages for example prompts
• OpenArt — Workflow and prompt examples
• PromptHero — Prompt database with generated examples

Common Issues and Fixes

Output ignores my prompt

• Increase cfg value (for SD 1.5/SDXL)
• Simplify the prompt — too many competing concepts confuse the model
• Check that the model matches your prompt style (anime model won't produce photorealism)

Specific element keeps appearing despite not being prompted

• Add it to your negative prompt
• Check if the base model has a strong bias (some fine-tuned models default to certain styles)

Prompt syntax not working

• Make sure you're using standard parentheses (), not Chinese parentheses
• Weight syntax is (word:1.3) — no spaces around the colon

Related Guides

• Text to Image — Basic generation workflow
• Embeddings Guide — Compressed prompt shortcuts
• LoRA Guide — Style and subject fine-tuning

Source References

• ComfyUI Text to Image tutorial
• ComfyUI workflow concept
• ComfyUI Getting Started with AI Image Generation