@Perchance, looking at the TextToImage plugin page again today the wrongness of the line ‘the same seed generates the exact same picture’ is sort of jarring since nothing else on the page is false nor misleading (that i kno of).
for those of you who dont know, the same seed generates a variety of similar pictures.
You must log in or register to comment.
Thanks! I’ve adjusted the wording here - let me know if you have better wording ideas or any other feedback. Also, if you find extreme differences with the same seed (like, entirely different image composition/structure), I’d love to see examples of that, since there might be ways to reduce it.
all same seed and prompt
This is usually the amount of variation I get. Here are a few examples. Same seed often is same color and same area where big objects are. the last one is just the word Prophetess to highlight how less tokens is less variation. i put them in the https://perchance.org/beautiful-people gallery.
spoiler
sameseed
sameseed
sameseed
sameseed
sameseed
Not this morning, but sometimes when moving GuidanceScale up or down on the same seed there hits a transition point at a specific guidance scale with that image where the variation between sameseed images is much larger.
<3 times infinity
you’re honestly the best dev
How would you reword it? “The exact same seed and prompt and negative prompt produces highly-similar images.” Something like that?
“The same seed and prompt generated multiple times results in nearly identical images.” maybe
Yeah. Just the way you wrote your post it sounded like some heinous, egregious sin, that the write of that page is lying to us or something.
It’s not 100% accurate, but an understandable simplification–in most cases the differences are incredibly minor. They could edit it to be more accurate, fair enough. But it’s not as evil or outrageous as it seemed reading your post 😅
you are so right. im going to only have my documentation be true sometimes now.
Also ironically in my mind I hadn’t even considered it being an ‘on purpose misleading’ until adcom mentioned it. My initial thought was Perchance used to work like that or was initially intended to and later the dev put various filters in which caused seed now to not always be identical. I thought it some innocent thing.
I am surprised by this post at how acceptable false documentation is and that things that are actually wrong are valid to put in as documentation. You won’t ever see me doing that unless it is accidental; in which case I 100% appreciate being made aware of it.
Perchance doesn’t really have documentation, I wouldn’t even call it that 😅 By no means is there any exhaustive documentation on pretty much anything to do with Perchance. Not written by the dev anyway. It’s all very temp and WIP and partial and incomplete.
I’ve made my own documentation for it, to my own standards. But haven’t touched things like AI generation. I would just send people to the page of whatever model/tech I’m using and let them do their own research there, instead of trying to cover everything myself in a document.
I think this is just how Stable Diffusion works. There’s always some “noise” to the system, even using the same seed.
As I said, bringing it up is totally fine. And they should amend it to be more accurate. And looks like they have, from the comment they left here.
I didn’t say it’s “valid to put in as documentation.” Just that I know what happened. It happened because a) the dev is not a documentation writer, b) is making this platform up as they go along (I’m sure they’d agree), and that’s their passion, not writing about it, and c) they probably wrote it in a hurry so they could move on to something else that interested them more and it was good enough so they called it a day. Oh, and it’s not really to the level of “documentation” of the AI generator; I don’t think that was the intent necessarily.
This isn’t a professional outfit, know what I mean? 😅 So basically… these things happen. 🤷 Also… yes, helping them pick up on these issues is good; they just need our help to do that.
Don’t read my response as “there’s nothing wrong with what you pointed out.” But responding to the idea that it was written to be “false and misleading.” It wasn’t written to be false and misleading, it just turned out that way. 😅 Like a typo in a book wasn’t put there maliciously, it just wound up being there, and the process of editors and proofreaders it went through didn’t pick up on it before now. Nothing on perchance has been through editors and proofreaders even–so you’re going to see mistakes like this. That were not maliciously or purposefully false or misleading. They’re just simple mistakes.
On top of that, maybe that wasn’t what you intended to come across, but just the wording made it sound accusatory like that. So naturally, if that perceived accusation is not true, you’re going to see some defense against such an accusation. I think that’s all that’s going on here.
That’s what I was saying about “inaccurate.” That word doesn’t have any connotation of wrongdoing or malintent. “False” and “misleading” do, however. See what I mean? If not that’s fine. Just explaining how what you thought I was saying about acceptability of the problem isn’t accurate either. 😜
Perchance dev is correct here Allo ;
the same seed will generate the exact same picture.
If you see variety , it will be due to factors outside the SD model. That stuff happens.
But it’s good that you fact check stuff.
feel free to delve in to it and form your own conclusion. One observation is the amount of variation ‘seems’ related loosely to amount of tokens.
It doesn’t. Noticed it long ago and some of us delved in to it ina post and VioneT explained it has to do with noise variations. The off part to me is the Perchance documentation saying a false thing, since that’s highly unusual and jarring with the rest of it; hence mentioning it
I get it. I hope you don’t interpret this as arguing against results etc.
What I want to say is ,
If implemented correctly , same seed does give the same result for output for a given prompt.
If there is variation , then something in the pipeline must be approximating things.
This may be good (for performance) , or it may be bad.
You are 100% correct in highlighting this issue to the dev.
Though its not a legal document , or a science paper.
Just a guide to explain seeds to newbies.
Omitting non-essential information , for the sake of making the concept clearer , can be good too.
3Tho I do not think this is one of those areas.
Long ago I actually made a generator that was based off this concept of same seed = same image. After a decently long build time, it turned out that same seed does NOT equal same image and was an unforseen wrecking of that design concept; not just unforseen, but I had read this same sentence I am highlighting now and it is specifically this sentence that made it sound good to go but as an unexpected failure with wasted effort.
It can be explained to newbies without causing harm. But right now it isn’t ‘ommitting nonessential info’. It is portraying something incorrectly and, if anyone uses the info for a generator using the concept of Same Seed = Same Image, they get pwned by the statement having been unexpectedly false.
I do not know anywhere else in Perchance where something stated as true is actually false and if you use that info assuming it’s true you get pwned.
Fair enough
Deleted by creator.
