Prompt weighting syntax: (word:1.2), [word], and how A1111 and ComfyUI differ

Both A1111 and ComfyUI let you raise or lower attention on part of a prompt. A1111 syntax - (red scarf) multiplies attention by 1.1. - ((red scarf)) is 1.1 x 1.1 = 1.21. - (red scarf:1.3) sets it explicitly. Prefer this; it is readable. - [red scarf] divides by 1.1. - (red scarf:0.7) de-emphasises explicitly. - A literal bracket in the text needs escaping: \( and \). - [cat:dog:0.5] is prompt editing: "cat" for the first half of the steps, then "dog". ComfyUI ComfyUI accepts (word:1.2) as well, but by default it does not rescale the whole prompt afterward the way A1111 does, so the same weight can feel stronger in ComfyUI. If you move a prompt between them, start by pulling weights toward 1.0. Custom nodes exist that emulate A1111's weighting if you need parity. Sane ranges - 0.8-1.3 for nudges. - Above about 1.5 the concept starts to dominate and distort: textures repeat, colours bleed into everything. - If you need 1.6 to get something to appear, the model probably does not know the concept well. Use a reference image, a LoRA, or inpaint it in. Method Same seed, same prompt, the target term at 1.0, 1.2, 1.4. Post the strip. The point where it starts to break is what other people need to know.