No.69
>>65Maybe I need to write a guide for the stuff I've learned somewhere. It's hard to find a place for it on kissu because /megu/ is supposed to be volatile like /secret/, but I'm focused on sex stuff so having a thread of nothing but spoiler images feels weird
No.70
>>69We could make an undying AI/General H-stuff guide thread here, seems useful enough towards the goal of enabling people to produce more ero content for /megu/
No.72
huh i was expecting moral corruption not demonic
No.79
>>65I experimented with making a 'storied' image series once, and accomplished it by using a single seed and just making gradual sequential changes to the prompt in the direction I wanted it to go, keeping changes that worked well and undoing those that didn't.
No.87
>>79How does the seed function work anyway?
No.90
>>87keep the seed the same = reproducible results.
changing the sampler gets you pretty similar pictures (less similar if it's an ancestral sampler), same with changing the cfg scale or the number of steps. in my experience changing the prompt slightly means you see parts of the image change while other parts stay the same. typically if you have a high-contrast edge somewhere in one prompt's image, that edge will be preserved in location and orientation while being 'repurposed' for a different object
No.184
>>87Long story short, Stable Diffusion is basically just an advanced denoising algorithm that reinforces patterns in inputted images, with the desired patterns being defined by the model and the prompt - the trick though is that rather than inputting an existing image of what you want, you instead just provide it with some white noise so that it can 'find' patterns where there weren't actually any to begin with. The seed is the label for the particular white-noise image you input, and so defines the starting point for each generation. Making small changes to the prompt while keeping the seed the same will often (although not necessarily) only change the output slightly, whereas changing the seed is like totally rolling the dice again. Also, as a general rule, I find that with a longer prompt you will generally get less variation, both between seeds and when making small changes to the prompt, than with short prompts, which can be either a good thing or a bad thing.
No.198
I still don't get how I'd find an ideal seed...
No.199
>>198Just try a bunch of random seeds until you get something that looks close to what you want, then stick with that one. Also, if you have previously generated an image similar to what you are currently after then it is sometimes worth trying with the same seed.
No.200
The ideal seed is the one that comes from the penis of the one you love
No.709
>>704Love those girls. I wish I could recruit them to be spies for me and then when their missions are done. They could crawl in the bed with me for a wonderful night of fun.
No.1309
>>1308Hmm... these colors are definitely not right. Seems like something weird is going on.
Well, and it helps to do a upscale to give better detail, but hmm... those colors. What was the model name exactly?
No.1311
>>1309That was under the Kissumono mix