Stable Diffusion Prompt Weight Converter
Paste a prompt written for one Stable Diffusion UI and get the same weights in another UI's syntax. The converter reads (word:1.2), ((word)), [word], word+, {word} and 1.2::word ::, works out each term's real weight, and rewrites it in the syntax you pick.
FreeNo signupRuns in your browser
| Weighted term | Real weight |
|---|
Weights only tune a prompt that is already good. If you want a strong starting point, the prompt directory has free prompts you can paste and adapt.
Browse free copy-ready prompts →Prompt weight syntax cheat sheet
Every Stable Diffusion front end lets you push a term up or down, but each one spells it differently and uses a different step size.
| Interface | Stronger | Weaker | Exact weight | Step |
|---|---|---|---|---|
| AUTOMATIC1111 / Forge | (word) | [word] | (word:1.3) | ×1.1 / ÷1.1 |
| ComfyUI | (word) | (word:0.9) | (word:1.3) | ×1.1 |
| InvokeAI (Compel) | word+ | word- | (word)1.3 | ×1.1 / ×0.9 |
| NovelAI | {word} | [word] | 1.3::word :: (V4 and later) | ×1.05 / ÷1.05 |
Nesting multiplies. ((word)) in A1111 is 1.1 × 1.1 = 1.21, and {{{word}}} in NovelAI is 1.05³ ≈ 1.16. That is why a prompt copied from one tool to another often looks wrong: the brackets survive the paste, but they no longer mean the same number.
How to use the converter
- Paste your prompt. The example shows nested parentheses, an explicit weight, square brackets and a LoRA tag.
- Leave Written for on auto-detect, or set it yourself if the guess is wrong. Square brackets mean different things in A1111, ComfyUI and NovelAI, so this choice changes the result.
- Pick the target syntax and copy the output. The table underneath lists the real weight of every emphasized term, which is a quick way to spot one that is far too high.
LoRA and other <...> tags are passed through untouched, and so are A1111 prompt-editing blocks such as [cat:dog:0.5] and [cat|dog], because those control scheduling and not weight.
Why the same weight looks stronger in ComfyUI
A1111 and ComfyUI accept the same (word:1.2) syntax but do different math with it. A1111 scales the token's embedding and then rescales the whole prompt so its average stays where it was, which softens the effect. ComfyUI moves the token away from the empty-prompt embedding and does not rescale, so the same number lands harder.
Nesting differs as well. In A1111 an explicit weight multiplies with the parentheses around it, so ((cat:1.2)) is 1.32. In ComfyUI the explicit weight replaces them, so the same text is 1.2. The converter reads each prompt by its own UI's rules.
The converter carries the resulting weights across unchanged, rounded to two decimals, since that is the faithful translation. If an A1111 prompt looks overcooked in ComfyUI, pull the highest weights toward 1.0, or use a custom text-encode node that offers A1111-style weight interpretation. There is more detail in our guide to Stable Diffusion prompt weights.
When weights are the wrong tool
- Word order comes first. In SD 1.5 and SDXL, terms near the start of a prompt tend to carry more influence. Move the important thing forward before you reach for 1.4.
- Negative prompts remove things better than low weights do. To get rid of something, list it in the negative prompt. The negative prompt generator builds one for your model family.
- Weight support for Flux and other T5-based models varies by tool. Say what matters in a full sentence first. The image prompt generator can write the same idea as tags or as natural language.
Frequently asked questions
What does (word:1.2) mean in Stable Diffusion?
It multiplies the weight of that word by 1.2, so it pulls harder on the image than the rest of the prompt. Values below 1, such as (word:0.8), reduce its influence. The same syntax works in AUTOMATIC1111, Forge and ComfyUI.
What is the difference between ( ) and [ ] in AUTOMATIC1111?
Each pair of parentheses multiplies a term's weight by 1.1 and each pair of square brackets divides it by 1.1. They stack, so ((word)) is 1.21 and [[word]] is about 0.83. Square brackets that end in a colon and a number, such as [cat:dog:0.5], or that contain a pipe, such as [cat|dog], are prompt editing and alternation, not weights.
Do AUTOMATIC1111 prompt weights work in ComfyUI?
Partly. ComfyUI reads (word:1.2) and (word) the same way, but it does not treat square brackets as de-emphasis, so [word] has to become (word:0.91). ComfyUI also applies weights without A1111's renormalization step, so the same number usually looks stronger, and an explicit weight inside extra parentheses replaces them instead of multiplying.
How do I convert NovelAI curly braces to A1111 weights?
Each pair of curly braces in NovelAI multiplies the weight by 1.05 and each pair of square brackets divides it by 1.05. Three pairs of braces is 1.05 x 1.05 x 1.05, about 1.16, which becomes (word:1.16) in A1111. Paste the prompt above and the converter does the math for every term.
What is a safe range for prompt weights?
A common rule of thumb is to stay between about 0.5 and 1.5. Higher weights tend to produce harsh colors, repeated objects and distorted anatomy, and how soon that happens depends on the checkpoint and the interface. If a term needs more than 1.5 to show up, move it earlier in the prompt or reword it instead.
Do prompt weights work with Flux?
It depends on the tool. Weight syntax was designed around the CLIP text encoders in SD 1.5 and SDXL. Some interfaces also apply weights to Flux's T5 encoder and others bypass the syntax entirely, and the effect is less predictable either way. Describe what matters in plain language first, then test a weight in your own setup.
