How ComfyUI prompts work
- Positive and negative prompts are separate CLIP Text Encode nodes wired into the sampler — there’s no single prompt box.
- Numeric weighting works: (token:1.2). Skip Automatic1111’s bare nested ( ) / [ ] emphasis — ComfyUI doesn’t interpret it the same way.
- LoRAs load via a Load LoRA node (separate model and CLIP strength), not inline <lora:…> text.
- Use Conditioning (Concat) nodes to isolate concepts — the equivalent of A1111’s BREAK.
Example ComfyUI prompt
masterpiece, best quality, a lighthouse on a storm-lashed cliff, dramatic clouds, (volumetric light:1.2), cinematic, ultra-detailed
negative: worst quality, low quality, blurry, deformed
Build ComfyUI prompts faster with Prompt Builder
- Compose the positive and negative prompts with reusable snippets, then paste into your CLIP Text Encode nodes.
- Drag in a ComfyUI PNG to recover its embedded prompts — no node spelunking.
- One-key (token:1.2) weighting without counting parentheses.
- Save and share full prompt recipes with the community.
Frequently asked questions
How are ComfyUI prompts different from Automatic1111?▾
Both take tags and (token:1.2) weights, but ComfyUI uses separate nodes for positive/negative and for LoRAs, and skips A1111’s nested-bracket emphasis.
Can I import a prompt from a ComfyUI image?▾
Yes — drag the PNG into Prompt Builder and it reads the embedded ComfyUI workflow metadata to recover the prompts.
Does (token:1.2) weighting work in ComfyUI?▾
Yes, the numeric form works. Use a Load LoRA node for LoRAs rather than inline <lora:…> text.
Keep exploring
- Prompt weighting & LoRA syntax guide
- Prompt glossary
- Other models: SDXL · Flux · Midjourney · Pony Diffusion · Illustrious · Stable Diffusion · Automatic1111