Attention reduction to a prompt has minimal effect

[giteeeeee]

Example:

With the example, I want the female to look like only 10% of the old, which can't be done. This is also the case with other prompts with attention reduction as low as 0.001.

Any ideas?

[ghost]

use "portrait of a female:9 AND portrait of an old female:1"

edit: prompt weight must equal to 1, so please read further down and dont use my above example as it is

Havent tested it yet but SD uses some kind of hack that differentiates it from other diffusers like disco diffusion. There is an episode on the youtube channel "Computerphile" about SD and why it is so different. Your weighing might mean nothing if the hack that SD uses directly points to the "old woman" latent space.

Or maybe I am completely wrong and it is a bug

[giteeeeee]

Thanks for the input. And yes I've been wondering if this is how sd suppose to be or an actual bug.

[ClashSAN]

In various prompt blending or prompt interpolations scripts, they only use a weight of 0-1.

portrait of a female:0.9 AND portrait of an old female:0.1

[giteeeeee]

In various prompt blending or prompt interpolations scripts, they only use a weight of 0-1.

portrait of a female:0.9 AND portrait of an old female:0.1

Thanks for answering. Anywhere I can read about this "a AND b" syntax? Don't think it's in the wiki.

[ClashSAN]

also supports weights for prompts: a cat :1.2 AND a dog AND a penguin :2.2

its in the readme and in some comments in the code c26732f

[R-N]

I know this already has an accepted answer but I'm finding "AND" hard to use, specially with styles.csv or a lot of prompts. I can hardly make it blend like I wanted. Would be easier if there were grouping. But anyway, here's an alternative. You can use the prompt editing/scheduler syntax [from:to:when].

For something major that needs to take shape, use from with low when. For example, [dress::0.2]. Something that's already in the picture will likely stay in the picture, and if it keeps being applied, then it won't be editable. Hence the low when.

For something minor that edits/alters something, use to. For example, [:clothes removed:0.3]. Some prompts has strong effect, some other has weak effect. The when should be low (early) or high (late) depending on it.

You can also combine the two to put something in the middle of rendering. For example, [[:sobbing:0.3]::0.5] will be added from step 30% to 50%. I found this somehow necessary. When applied near the end, tears won't be a separate object and instead alters the eye shape. Meanwhile if it's applied in the beginning, it affects the picture a lot but may hardly take shape, so instead it adds some obscure skin/eye colored bulge.

If it's not strong enough, you can extend the application steps through when, or you can also add weights to it with (prompt: weight).