[Feature Request] Allow multiple scale values for one job

[twiffy]

I'd like to be able to one-click start a job that runs a prompt over multiple scale values.

One option is to allow a single interval of scale values, e.g. using a two-node slider bar. But there's also a use case for allowing more complex sets. In one of my experiments, I noticed that the best results happen at scale values 9, 11, 13, 13.5, and 14. (Very random, I have no idea why.) So I could imagine having an option in Settings to toggle between several modes of specifying scale. Maybe one option is a one-node slider, and the other option is a text box that allows notation like

9, 11, 13-14

If implemented, to me the most intuitive behavior would be processing the batch size and batch count for each scale value in the specified set, so that the total number of generated images would be (batch size) * (batch count) * | scale values |.

[twiffy]

Now that I think about it, there are probably some other parameters for which this would be useful, e.g. Denoising Strength in img2img.

[EyeDeck]

Use the X/Y Plot script for this:

Input

Output

White duck in a sakura forest. Concept Art. Night scene. Cityscape. Detailed, intricate. Oil on canvas by (J.M.W. Turner). Steps: 100, Sampler: Euler, CFG scale: 4.0, Seed: 2661864062, Size: 512x640, Model hash: 7460a6fa, Denoising strength: 0.3, Mask blur: 4
(I hastily changed another prompt and left some irrelevant stuff in there, but good enough)

[Naeaeaeaeae]

I dont see a denoising Option. Was it only in one version available?

Edit: Sry i thought Denoising is also available with txt2img

[dermesut]

well, i did write a script for... something similar... and a little bit more.
basically, you choose one of those three modes, define the min/max via sliders and the script subdivides the range by taking the "ImgCount" slider into acount. say, you want to "travel" the cfg-scale between 0 and 10 und your "ImgCount" is set to 5, you'd get images with the cfg-scales of 2, 4, 6, 8 and 10.
same goes for denoise-strength.
seed-travelling is simply adding +1 to the starting seed. no sliders for that one.

additionally, i built in a "blind-folder"-option.
without being given a specific path, it grabs, randomly, an image out of a folder, that is hard-coded within the script. it's very unlikely that this path would make sense on any other system than mine, so i'd suggest to put in a specific path on your own into the text-field. if there is more than one image in the folder, it would randomly pick one of them. the web-ui frame can be left empty.
if the web-ui frame is not left empty, that is, if you put in an image, as you would usually do in img2img, the "blind-folder" option does nothing.

and: there is the checkbox, which includes your init pic into your series of rendered pictures.

and: there is the "Use Lexica" checkbox. Lexica recently opened up an api and provides a set of 50 items per call.
the script only uses the "prompt"-value of those by randomly choosing one out of those 50 and adding them to your prompt.
so, there a few combinations:

• your prompt is empty, "use lexica" is deactivated -> your prompt gets filled with the prompt "randomness" (not very clever, i know).
• your prompt is not empty, "use lexica" is deactivated -> your prompt gets used, as one would expect.
• your prompt is empty, "use lexica" is activated, you didn't put any "clues" in -> your prompt becomes the result of the scripts's efforts to look for the word "randomness" within lexica, which gives back 50 sets that "are somehow associated" with the word "randomness". randomly, one of those 50 prompts is being chosen and becomes your prompt.
• your prompt is empty, "use lexica" is activated, you put "clues" in -> your prompt becomes the result of the scripts's efforts to look for the words that you put into "clues", within lexica, which gives back 50 sets, that "are somehow associated" with your clues. randomly, one of those 50 prompts is being chosen and becomes your prompt.
• your prompt is not empty, "use lexica" is activated, you put "clues" in -> your prompt becomes the result of the scripts's efforts to look for the words that you put into "clues", within lexica, which gives back 50 sets, that "are somehow associated" with your clues. randomly, one of those 50 prompts is being chosen and is being added to your prompt.

finally, the debug checkbox is just for my testing, making some things visible.

that's it, basically.
i have not thouroughly checked, if or how the naming-features of web-ui are being fulfilled, wasn't that much of a concern until now.
this is a wip, after all.

that said, i am no real programmer, so, expect spaghetti-code.
basically the loopback-script was the source of everything.

on that point: i have not managed to avoid the seed being +1 on each image count.
whenever i tried it, the "travels" on the denoiser or cfg-part seem to kind of "degenerate", so i kept it that way, that even though one is only travelling the, say, denoiser-road, the seed still is +1 at each step.
what i would have wanted: same seed for every image, while only changing denoiser (or cfg).
can't seem to make that happen, without images degenerating over time.

finally:
i have never used github really, just recently been trying to get used to it. i seem to have managed to git pull (within in pycharm, which is what i use), to keep this repo locally up to date. but i never "pushed something up", or whatever that's called.
sooo, i'm not sure, how i would let you have that script. i'm a bit shy to break something on this guthub-repo, if i'd try and "push" something up.

any ideas? maybe i could just paste the code into this comment-text-field, and you generate your own python-file and place it into the scripts-folder of your own local installation? what do you think?