took a few tries but im impressed with the result, i was trying to create a young fat latina and this was the best pic by far

i wanna know if there is any specific or general order on how i should strtucture the prompt

Its definitely gotten better, a bit more responsive and fun - made these with a bit of post-production

>>26794 Wow. How’d you get a series of progression? And what do you mean by post production?

...ok really dumb question, but how are people using hyperfusion, when all i can see is hyperfusion finetune?

holy shit it looks like you can make gifs now, anyone have a guide how to do it? I don't see the option anywhere.

>>25252 >>26094 Lovely

Spent a lot of time on this site a few months ago. It’s a fun novelty, and easy to learn. Would recommend

It's been fun recently with the new updates they've had.

https://www.deviantart.com/maxweight/gallery

>>26794 What's your process like? How do you keep the faces consistent while doing such different bodies?

>>27099 >>26865 The nutshell with these is a sort of 'feedback loop' - I start with a skinny chick photo thats uploaded, that "locks" the pose and then add "skinny" prompts & prompt the clothing. Get a good seed and result and start adding weight to the prompts... eg chubby, flabby, all the usual. StufferAi will fatten up the skinny chick base photo a little bit ... then I upload the fattened up image and repeat the process. when rendering full body the Stuffer faces suck - I upload a cropped close up of the crappy full body face for source and it'll generate a good face which is easily added - lotta prompts like chubby face, underchin, etc Used photoshop to quickly composit the faces and laid them out as a progression. Its not the easiest way to do things but gives fun results up to a certain point

>>27236 Got you. The only thing I don’t understand is uploading a face as a source. Where is this ‘source’ option? As in control net++?

>>27761 When did they switch to webp files?

I'm just going to post some sequences I made/pics I came across that I liked. That alright?

>>27775 ...I don't know how to post these in the right order, soz.

>>27771 Since probably about a week ago. I've been following the keywords guide. And I just wanted to share all the adjectives I've found in order of size. I'm looking to make basically a script that will generate a sequence of keywords you can give to make a corresponding weight gain sequence: thin, curvy, chubby, fat, big, large, {I need an intermediary here, because there's a jump to the next keyword}, huge, gigantic, {another jump}, colossal, hyper I've also been trying to get that "really full" look, and a lot of times "full, stuffed, bloated" doesn't cut it. I've found "torpedo belly, outie_navel" does it--sometimes I'll have to add a "hyper pregnant." Also, how do I make the images here post in order?

I was considering donating but the inability to make a one-time donation really fucking sucks.

Can the program recognize anthro dragonesses? Cause if so can anyone make a ussbbw/sbbw/bbw (dealer's choice) pregnant/hyper pregnant dragoness? Bonus points if you can get lesbian dragonesses that big

Any idea when the chat will release? It'll be interesting seeing an AI chat dedicated to weight gain/stuffing

Does anyone know how to download the videos that have been going up? I can apparently grab the thumbnail but the video itself won't let me grab it.

>>30502 You can inspect element, then filter for .mp4, then open that link you find and you can download it.

Guy who originally made the thread here, even if the ID system doesn't say as much thanks to swapping VPN providers and quite a bit of time passing; I went back and spent an entire night of migraine medicine-induced insomnia dicking around with the site and genning a ridiculous amount of fatties, specifically using only the Hyperfusion v9 model for drawn ones since I don't much like the photo models still, and I made some interesting discoveries I'd like to share with anyone here who might be interested. Namely, I discovered that despite the fact that 99% of the top posts on the site using Hyperfusion all seem to use "c.cu" and "blushyspicy" as the artist tags for vaguely anime-style art, this thing can and will replicate more western/cartoon style artists like woot's just fine as well (images 1 and 4). More notably, it was able to replicate some less-well-known fat artists when I tagged them as well, such as the thick lined style of joekie3wl in the second image, and actually producing a full-on 3D style render to replicate ShubiJubi's stuff in the third image which surprised the hell out of me, as I expected it to just get confused and spit out something more generic in 2D like it did with other artists I plugged in that it didn't have enough or any data on. And to the credit of the people actually working on this thing, it took me quite a bit to actually find artists obscure enough to throw it off, I was literally plugging in random names off twitter and deviantart whose styles I found appealing mixed with scenarios I liked and seeing how they turned out, very rarely did it fail to actually translate the style properly. For an example of good shit it did that I personally like a lot, check image #5 that uses mentalcrash's artstyle, no points for guessing what the OC donut steel character was vaguely inspired by. Continuing my rambling shit in a second post because I need to add more images as examples to go with what I'm going on about, there's a (hopefully) good reason for it so please just roll with it I guess

>>31410 However, all of this pales in comparison to what I discovered toward the tail end of my session of tinkering with it, where my tired brain opted to try something incredibly stupid on a whim and it actually worked. See, by default, the image gen prompt works off a typical priority system where parentheses denote extra emphasis, i.e. if you put in "(huge tits)" the generator will put more emphasis on that in the image output, and doubly so if you write "((huge tits))" like so with doubled parentheses. You can even add in ":1.5" or similar to increase the priority even more, usually gives absurd proportions. That bit is normal and, as far as I know, pretty standard for image gen systems. What's NOT normal, however, is that this one seems to be able to recognize that if you put a specific phrase like "(Viewer POV - [insert action here minus the brackets, commas can be included to help organize info without it counting as separate tags within this description])" after everything else, it can and will read that as an extra description to modify the output accordingly as if whatever you described occurs in the image right after the initial generation occurs, and is then added to the final output. In other words, if you prompt an image of a fat girl with a huge belly, then tack on "(viewer pov - the viewer uses their hands to rub her belly from offscreen)", the generated image will include hands coming in from offscreen to do just that, complete with appropriate reaction from the character. This also extends to more complex and sexual actions too; You gen an image of a fat chick lounging on a couch and add a bit about her straddling and banging you, and boom, now you have an image of her riding you while still looking the way you specified previously, usually with a randomly-generated male body and dick visible in-frame under her to complete the scene properly. It'll read this line like an instruction set in something like chatGPT, rather than a tag list like the rest of the image prompt typically does, and it's insanely-flexible what you can do with it. Naturally the obvious use is to generate really good wank material, but you could probably make all kinds of insane things that aren't porn with it too if you're creative/insane enough. Here's some basic examples; First one is simple enough - generating brunette with poofy hair and a big gut, then doing a second gen involving the viewer rubbing said gut. Third is same thing but adding lactation to the mix. Second set is going from standard gen of mild slob NEET to full-on sex at record speed with just that one addition via the "viewer POV" instruction set. Granted, the effect might not be very clear through the limited info available in a post and a set of images with no real metadata, but still, I strongly recommend giving it a try on the website yourself and seeing what you can do. If absolutely nothing else, being able to radically-alter image outputs without having to rewrite most or all of the prompt is pretty nice, as is having the generator recognize a wide variety of styles that aren't just the same one or two generic styles that keep getting recycled in the top posts on the site (not a dig at anime to be perfectly-clear with my words, just a gripe about people being repetitive with their tastes in general) Spoilered the last 3 images for being a bit much in case people don't wanna see that shit unexpectedly

Some experimentation and a depressing number of hours of my life wasted later, I've found out a few things that seem to work: -As previously stated, it's best to just browse the top pictures, find something you like, and copy the prompt, rather than slowly iterate a prompt from scratch. -Set KSampler steps to 20. It defaults to 15 and while the bar looks like it's at 100%, it can go higher. -Everything but hyperfusion v9 (I think I've gotten my best results from epoch 13) produce some uncanny valley bullshit or is at least sloppier than most sloppa. Similarly, every scheduler besides Euler A(ncestral?) kinda sucked. -For controlnet, I found it worked well enough at 0.3 strength. At 0.5+ it often had too much influence over the final slop, doing things like overriding a tag like "hands on hips" by sticking in a third hand rubbing/grabbing the belly if the original image had something like that. Probably something easily fixed by choosing a better image to steal from. -Resolutions that don't include 1024 in one or both dimensions sometimes mess up the color/lineart sharpness. -The tag "hyper_thighs" at :2+ made EVERYTHING have stupid fucking :3 hamster faces, which I assume came from some furry-based data. -In general, any tag set to a value beyond :3 causes weird shit to start happening, although depending on your tolerance :4 is a good limit. At :5+, The Warp sets in, and turns you're waifu into abstract art. -From what I can tell, "hands_behind_back" works much better than "hands_behind_head" for hiding those strange fleshy tentacles at the ends of a character's arms. -"cityscape" seems to be tailored for giantess stuff, kind of surprising, but if you want regular urban chicks, just "city" should do. To the owner, or anyone else in the know: is there a way to add comments to prompts without interrupting the tags? Neither //text nor <text> seemed to have an effect, but it would be nice for more readable formatting.

>still can't gen shortstacks or women with anything less than massive titties (at least intentionally) >AI has filters in place against the usage of the words "shortstack", "goblin", "small breasts", "short" and pretty much any other words/combos that could conceivably be used to lead to genning underage shit with some effort, presumably to ward off pedos Not sure if I should be annoyed with pedophiles for ruining good things by being dipshits as always, or with the people running the AI for for being overly-cautious with the filtering/going back on their original promise to nix the limits after initially dialing them back, as you can see in the early posts in this very thread. I wouldn't be shocked if it turned out they had to put the limits back in place due to spam from the usual morons, nor would I be surprised if they just did it just because even if there was no such spam, pettier shit has indeed happened on the internet before. Mind you, I just want to generate short women who are also fat, along with women with the BBBXL body-type where they aren't especially-busty since it makes their bellies and asses all the more pronounced, which I consider to be some primo shit. I like big tits as much as the next guy, but some variety is nice, as is some height difference. If the devs could remove the obnoxious filtering again without the site being blasted by pedo bullshit, that'd be nice, but I'm not exactly holding my breath for obvious reasons knowing how people are.

>>31733 I will admit its a little frustrating trying to use "smother" and it scolding you for the word containing "mother". but its free image generating so i try not to complain

Stuffer ai is great ^^

Ive been cooking

>>30124 Also wanted to know. But since adding an AI chat will need their own interface incorporated on the site, perhaps it will take a while

>>31737 Fair enough, I don't mean to be ungrateful. I'm very glad to have this, as between being far too dumb to setup AI on my own system and lacking the funds to pay for a subscription to buy a setup on someone else's service, without it I'd have literally-nothing. Just wanted to point out the one and only issue I have, in the hopes it might be noticed and fixed. And funny enough, within about a week of me mentioning it here; >go back on again this evening and check top posts of the past 3 days >find a top post that looks damn good and isn't the usual generic anime stuff that usually floods the front (just not my taste, nothing against it) >notice it uses "small breasts" and "short" in the tags just fine (pic related, not my creation, it's the pic in question using the tags that I found) >try it on a whim on something terrible I genned not worth posting >now it works, and the filter no longer nags me like it thinks I'm some scum-sucking bastard trying to gen sick shit, rather than just a regular weirdo trying to gen cartoon fat chicks for free Coincidence or not, I'm happy either way

>>31883 yeah it changes occasionally. ive noticed it sometimes ignores certain words if it feels the prompt doesnt imply its for forbidden reasons like how it will on occasion allow you to use adjectives like "short" or "unhealthy" without blocking or hiding what you generate in the case of the later.

>>14465 Stuffer.ai is stuck at the "Generating" phase, is anyone experiencing that? Should I spoof my browser?

>>31896 It hasn’t worked on Mobile for me ever since the Star effect was put on the generate button.