Prompt building for ComfyUI & Automatic1111

ComfyUI and Automatic1111 (A1111 / Forge) read prompts in mostly the same language — but with important differences in weighting, LoRAs and how negatives work. Here's a practical guide, plus how to move prompts between them.

The shared basics

Both engines take a comma-separated list of tags or short phrases, processed by CLIP. Order matters a little — earlier tokens get slightly more weight — and both support explicit numeric weighting like(masterpiece:1.2). Start broad (subject, then style, then detail) and refine.

Negative prompts

A negative prompt lists what you want to avoid — worst quality, low quality, blurry, extra fingers. In Automatic1111 it's a dedicated box under the positive prompt. In ComfyUI it's a separate CLIP Text Encode node wired into the sampler's negative input.

Weighting differences

  • Automatic1111 / Forge: numeric (token:1.3), nested ( ) / [ ] emphasis, and inline <lora:name:0.8>.
  • ComfyUI: numeric (token:1.3) works, but skip the bare nested brackets; apply LoRAs with a Load LoRA node.

Full details in the prompt weighting & LoRA syntax guide.

Using BREAK and chunks

When two subjects swap attributes (the wrong character gets the red coat), separate them. A1111 uses theBREAK keyword; ComfyUI uses Conditioning (Concat) to join independently encoded chunks. Both reset CLIP's 75-token window so concepts stay distinct.

Moving prompts between engines

Most prompts copy across directly. When they don't, it's usually because of A1111-only bracket emphasis or inline LoRA tags that ComfyUI ignores. Prompt Builder helps here: drag in a PNG generated by either tool and it reads the embedded metadata to recover the exact positive and negative prompts, ready to tweak.

Wildcards for fast variation

Wildcards like {red|blue|green} or __lighting__ let you roll a fresh concrete prompt without rewriting it. Useful for batch exploration in both engines.

Frequently asked questions

How are prompts different in ComfyUI vs Automatic1111?

Both accept plain comma-separated tags and (token:1.2) weights. A1111 adds nested ( )/[ ] emphasis and inline <lora:…>; ComfyUI applies LoRAs and conditioning through nodes, and keeps positive and negative prompts as separate inputs.

What does BREAK do in a prompt?

BREAK ends the current 75-token CLIP chunk and starts a new one, isolating groups of concepts so they don’t bleed together. A1111 supports it directly; in ComfyUI you use Conditioning (Concat) nodes for the same effect.

Can I recover the prompt from a generated image?

Yes — drag a PNG made by Automatic1111 or ComfyUI into Prompt Builder and it reads the embedded metadata to recover the positive and negative prompts.

Build it without the syntax wrangling

Prompt Builder turns weighting, LoRA tags and negative prompts into one-click snippets and keyboard shortcuts — free on web, macOS, Windows and Linux.