>>1015 They're too expensive relative to the work they put in.

Friendly reminder that preg gens look nigh indistinguishable from belly inflation.

You have no excuze not to use 4chan.

Some classic and some unusual shapes:

>>18587 Same tired worn out dialogue from a tired worn out hack like yourself. What, get tired of starting shit in the slob threads?

>>18591 I'm glad it bothers you

>>18617 With your level of narcissism I doubt that. Let's not pretend now.

>>18622 content is content is content bb, whats the negativity about

>>18622 Also wtf is a "slob thread"?

Random attempts

>>11394 How are you getting that semi-realism, really nice effect here

>>18882 All really good but the last image is fucking incredible. I almost thought it was a render

>>18934 Cheers! The living inflatable flora and hyper fusion are the main factors, when I'm back at PC later I'll post the prompt bb

>>18934 here's the settings <lora:inflatable-v1.3:0.7> masterpiece, 1girl, ginger hair, round glassses, Huge breasts, Giant belly, Balloon belly,glossy skin, shiny skin, pink skin, transparent breasts, blue skin,, big breasts, huge ass <lora:hyperfusion_550k_128dim-LoCon_extracted-v7:0.7> Negative prompt: lowres, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, (worst quality, low quality), jpeg artifacts, signature, watermark, username, blurry, monochrome, motion lines, [3d], speech bubble, ((multiple navels)), (multiple Navels), (too many nipples), (malformed breasts),(malformed arms), (malformed limbs) ,ng_deepnegative_v1_75t Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2929667571, Size: 736x512, Model hash: ce30feb9a1, Model: realcartoon3d_v11, Denoising strength: 0.51, Clip skip: 2, NGMS: 6.66, Hires upscale: 2, Hires steps: 10, Hires upscaler: R-ESRGAN 4x+, Lora hashes: "inflatable-v1.3: 6049efd1740e, hyperfusion_550k_128dim-LoCon_extracted-v7: 66d481cf230a", TI hashes: "ng_deepnegative_v1_75t: 54e7e4826d53", Version: v1.6.0-2-g4afaaf8a

Hot take, realism is eh. Atleast for me. Anyway, big Frisks. Must've ingested something their body didn't like...

>>15053 Seconding.

Using locally-installed Stable Diffusion and BiggerGirls V2. I write, and I use AI generation as a way to play with ideas and get some images and inspiration back. So I haven't gotten deep into LORAs or anything more complicated. Just trying to get reasonable images out of SD, and then do some cleanup runs with a little inpainting or going out to GIMP. I'll drop a few here. I post to DA as joemonday.

>>21628 Sometimes SD's glitches can do fun things. In this one it looks like she's grabbing her stretchy belly skin and testing how much more room she has. Turned out to be one of my favorite images.

Few from the last wee while i liked

You don't need much. Big belly, pink nipples, maybe a hose and a pump.

You absolutely need to try the AI generals on 4chan.

>>22377 really looking forward to see your latest gens

>>22949 Will be posting some later but I have the majority of my stuff here, it's a range of fetishes but I'll post some blooms for you friend https://www.deviantart.com/eddoevil

I will never understand why some users will never try 4chan.

One has to wonder how much pudding you can pump into an attractive asian friend until she bursts like a waterballoon... Well, only one way to find out.

And of course, having one of your other cute friends test out your new tire pump...

Requesting real life-ification of pic related.

Pic related is an attempt on 4chan.

Can you run this through your magic machine and give us a couple of different styles, please?

Hello from PonyXL with InflationXL LoRa.

>>25824 yeah that's the one I use but it can be a little hard to get working with RL models sometimes. What model is that?

>>25848 I've had good luck with these two: - Pony Realism: https://civitai.com/models/372465/pony-realism - 7of9: https://civitai.com/models/341631/sevenof9ponyrealmix These are mainly merges imo, but the results start looking impressive, getting closer to realism with Pony prompting adherence and anatomy details

>>25897 Thank you! I'll have to give these a try later.

flux.dev is the tits. since there's no negative prompt, it's more about avoiding certain tokens and letting it fill in the blanks. this was nothing more than the GGUF and a boobs LoRA, with some suggestion about where the gas was going.

>>26967 agreed that flux is a huge step forward. Which boobs lora? A specific one or just one of them? I'd still like to know what your prompt was for this, mainly because I've found that trying to prompt one part of the body, namely the belly, results in it generating a larger person overall, even when you specifically tell it not to. The lack of a negative prompt is a problem, but it's still a major step forward overall.

>>27097 "belly" generates fat, and "pregnant" has too much weight on the pose and composition. avoid them. flux understands terms like body, stomach, abdomen, and flanks, and modifiers about size, like swollen, bulging, distended, spherical, stuffed, prominent, huge and very. it also has some action state recognition, like expanding, especially when there's a token aligned reason for it. each token has a weight based on the training corpus, and when prompting, we have to intuit those weights and balance them, usually with weaker supporting prompts. I don't modify attention, but I will blend tokens. it's like tuning a piano; each change you make is with respect to the final state you want to achieve. if you try to bring each string into tune individually, they'll all pull each other out of tune. only lora was big_boobs. don't remember the prompt, but if you play with the above, it should happen pretty easily. the only annoying thing was figuring out how to subjects them fully clothed. "dressed (in)" is as good as I've gotten.

>>26967 How do you even get it to work? ComfyUI is a pain in the ass to use so far and attempting to install the stuff for FLUX has completely broken it, even when following all the instructions.

>>27203 read this: https://comfyanonymous.github.io/ComfyUI_examples/flux/ it's got a slightly higher bar to first gen, but once you get how it works, the node/workflow system is significantly easier and more flexible than automatic1111. if you want to share your hardware, workflow, and errors, that might help.

>>27209 Managed to get it working, any tips/prompts/etc. to help folks get started making amazing stuff like this?

>>27212 prompt in concepts and short sentences for the content and regular tags for details, like SDXL. less is more. the aspect ratio and pose determine a lot about the gen, so squatting, kneeling, and sitting are most likely to give "room" for the subject of focus. all images were base flux.dev, the anime image used fca_style lora, and the villainess used big_boobs, which seems to be pretty good at enhancing large round things. the cfg range was 2.0-3.0, 16-20 steps - defaults for dev. experiment with samplers: https://civitai.com/articles/6582/flux1-dev-sampler-scheduler-comparison try the prompt guidelines I listed above. I don't want to set a pattern for prompting, because I'm not an expert (and there's enough sameface crap out there); the model is really just that good. get creative, post results!

>>27226 Do you need flux specific loras or will any work with it?

>>27226 Thank you, definitely getting some good results so far now that everything is working!

>>27227 yes, check civitai. there are like 50 new ones a day, and they're small compared to sd. >>27228 naisu. i've been using dpmpp_2m + sgm_uniform. ddim also gives great results but triples gen time. sometimes lowering the quality via prompting (a crappier camera, adding grain) improves the realism.