Dumping some old gens to kick the thread off. Some of these are from before I figured out ADetailer so apologies for the faces.

idk what I'm doing but I have just achieved a major aesthetic

>>36771 Should do Ran and Patchy with this style, I bet they'd be great.

>>36799 good idea. I went ahead and posted those in the touhou thread, here's something else

Local fatass gets retail job. Bonus video. >>36819 Absolutely loving Fat Power here, though I can't stop laughing at whatever happened to that girl in the beach tub.

Perchance

aqua is not enjoying it

Megachub office donut thief

really like how the two gyaru pics came out

Threw together a sequence of coffee shop chick from some recent gens >>37070 Gyarus really are excellent creatures. Good work on the two-tone hair!

trying out HunyuanVideo 1.5 I know this is amazing, but waiting 5 minutes for a short video like this kinda dampens my excitement the 480p model is faster of course but the output is too blurry and doesn't look good

here's another one first try: 20 steps and 'EasyCache' enabled (apparently this reduces the quality) - 270 seconds second try: 22 and EasyCache disabled - 580 seconds using hunyuanvideo1.5_720p_i2v_cfg_distilled_fp8_scaled workflow from here https://blog.comfy.org/p/hunyuanvideo-15-native-support

the first one is just comical prompt: "a girl is standing outdoors. she turns around to face the viewer and her belly and boobs jiggle"

my conclusion today is that it's not worth it to spend so much time with video generation I did try WAN 2.2 too and it was like 50% faster but I couldn't make it do what I intended and it also seems to have worse scene understanding.

Some fat Tifa.

>>37398 I also tried WAN 2.2 for non-porn stuff and couldn't get anything aside from blurry analog-horror-esque twitchy ghost crap, but that may have been me not giving enough time for rendering. How long/how many frames did you go for? Either way the Kyoko in >>37391 is fantastic (esp. that first one)!

>>37554 >How long/how many frames did you go for? 4 or 5 seconds I wouldn't say that the results weren't usable, but at least it took way too long to get there.

>>37557 Ahh I see! Even that vid does seem to be at a higher fps than I was using, maybe I should go back & try again.

>>37687 the default is 24fps and I don't think you can change it

>>37680 Could you please post the full prompt you used to make that Rin image?

>>37740 1girl, solo, (feet out of frame), tohsaka rin, booth seating, sitting, black thighhighs, black skirt, thick thighs, overweight, midriff peek, blush, looking at viewer, (simple background:0.8), masterpiece, best quality, amazing quality, (magentapeel, rkrk12:0.6) + the hyper girls lora probably

>>37742 im running out of ideas for generations, iveen trying to get into multi-character generation in txt2img, with mixed results. It seems they are either eating each other out, or clearly separated, or the lroas get mixed. For sex, it seems only doggy style works.. For some reason i hardly ever get proper eating results, and weight gain sequences simply hardly ever work. Anyone knows how to make img2img's inpaint or sketch mode work properly? i can sort of toy around it, and make some changes, but it takes a lot of trial and error every time ideally id want a method i could replicate easily, keep style sort of consistent without getting mixed loras

"my little sister can't be this blobby"

>>38192 you need regional prompting to reliably generate multiple characters, or it's a gamble if the arrangement works out. if you want to try anyway make sure to use the "2girls" tag and don't put "solo" in the prompt. eating isn't that problematic in my experience but make sure you're using a good model and not some old SD1.5. weight gain sequences can hardly be generated in one shot, though some loras can help with that. I'd rather suggest genning the pics separately and putting them together in photoshop. anyway here's some new stuff

What model is best for making fatties right out of the box with no loras required to make them look good?

>>38879 i'm sure there's a checkpoint merge somewhere that does BBW out-of-the-box but why are you trying to avoid loras?

>>38881 Loras for fatties often change other behaviors of a model so sometimes it can be annoying to tell if I'm having trouble getting good results because it's not something my model can do or if it's actually just the Lora causing the problems.

>>38904 that's true I recommend picking a popular lora, then you can be sure it's not ruining the quality (too much). the hyper girls lora works well for me.