Platforms differ widely on weight syntax. Automatic1111 uses parentheses, Midjourney uses double colons — mixing them is a common bug.
Three platforms compared
| Platform | Increase | Decrease | Explicit value |
|---|---|---|---|
| Automatic1111 / Forge | (token) | [token] | (token:1.3) |
| ComfyUI | (token:1.2) | (token:0.8) | (token:1.5) |
| Midjourney | token::2 | token::-1 | red car::3 blue::-0.5 |
| Flux | no weight syntax | — | use natural language |
Automatic1111 / Forge details
- (token) = ×1.1
- ((token)) = ×1.21
- [token] = ÷1.1
- Explicit: (token:1.3)
- Same syntax applies to negative prompts
Midjourney double colon
Midjourney splits prompts into weighted segments via "::":
red car::2 blue background::1 mountains::0.5
Negative weights act as anti-weights:
forest::2 trees::-0.5
Forest yes, but suppress prominent tree shapes.
Flux: use natural language
Flux ignores weight syntax. Use natural language:
- "a strikingly red car against a faintly blue background"
- "very", "extremely", "strikingly" as intensifiers
- Put important items first in the sentence
Gotchas
Using A1111 syntax in Midjourney
(token:1.3) parses as nothing in MJ. Use token::2.
Weight 2.0+ overcooks
Stay in 0.5–1.5.