/bbwdraw/ - BBW Drawn

Deviantart, 3D, scribbles

BBW-Chan is supported by simple text or static image ads from a-ads.com.
Please consider whitelisting us in your adblocker if you enjoy your stay!


Mode: Reply
Name
Subject
Message

Max message length: 9999

Files

Max file size: 10.00 MB

Max files: 6

Captcha
E-mail
Password

(used to delete files and postings)

Misc

READ THE *RULES* BEFORE YOU POST!

[ / / ]

for the first time in my life decided to just put fat anime girl in prompt with anything else. All the pictures kinda look the same, so here u can have a couple of examples of absolute lame
anyone got the wgt for bigger girls
test
(54.84 KB 512x512 1.jpg)
(51.32 KB 512x512 2.jpg)
(45.17 KB 512x512 4.jpg)
(43.69 KB 512x512 8.jpg)
Is it OK to try with more realistic prompts?
(855.63 KB 712x712 7.png)
(832.82 KB 712x712 23.png)
(932.15 KB 712x712 8.png)
(896.29 KB 712x712 9.png)
(967.57 KB 712x712 25.png)
(803.97 KB 712x712 26.png)
More if you're interested
>>137448 >>137449 >>137449 wtf ya using to get those results and whats the mix i got this with just the bigger girls model
would anyone be willing to generate more gray wolf?
(422.23 KB 512x768 tmp5n0xr_ax.png)
(448.14 KB 512x768 tmp5r0qr9ne.png)
(347.97 KB 512x768 tmphpz7zqk5.png)
(527.79 KB 640x768 tmp6r_c9axa.png)
(338.47 KB 512x640 tmpj9qjfurz.png)
(278.95 KB 512x640 tmpd6_rtv5u.png)
Year of chubby bunny
Is there any way that you can make these cool? Like can you give them dongs or something?
>>137503 Ye you can, but you'll have to go to /ee/ with that stuff. If you need prompts for that check AIbooru, there's enough stuff like that
>>137502 how do you get such consistency in the design of the pink cat? could you share an example of some prompts or parameters?
>>137473 what model did you use
i know this isnt exactly a request thread, but if someone could generate some of vtubers buffpup or shylily i would be forever grateful. (rough is ok since i assume you guys dont have trained models for them)
>>137626 i bow down to you, thank you. their designs are so complicated id love if you shared the prompts. even if they take 20 minutes on my pc. or if you can manage another batch with them larger. sorry if im asking for too much haha.
>>137653 I actually like this. What prompts and model did you use?
>>137652 >>137653 Wtf these are genuinely amazing
(492.18 KB 512x768 tmpnk7c0eff.png)
(487.21 KB 512x768 tmp6d0i9g4l.png)
(532.90 KB 640x768 tmpbdlgug2d.png)
(634.12 KB 512x768 tmp0tspuh_1.png)
(364.50 KB 512x768 tmpg4ktja6b.png)
>>137462 The bigger girls model is cool but it makes way better results when you merge it with another model. I merged with AnythingV4 I also trained it a little using the last Ben's dreambooth colab. It got an update that lets you train using training images with captions, which really helps the quality and also lets you train specific tags e.g. fats, hips, T&A, futa, feet, etc Also use DPM++ 2S a Karras sampler if you have the option, it is about as fast as euler but gives better details and can also do better hands. 10-20 steps for roughs and 40-60 for real pics >>137502 Kino >>137515 Idk what the other guy used in his prompts but I used "nyatasha nyanners, light pink hair, pink cat ears, twintails, purple eyes, hair ribbon" You'll have to inpaint to get black hair ribbons cause they usually come out pink or purple regardless of prompt and you can replace "twintails" with whatever hair you want
(784.44 KB 512x800 651.png)
(891.28 KB 512x800 644.png)
(784.60 KB 512x800 671.png)
(799.27 KB 512x712 963.png)
(1.12 MB 640x976 223.png)
(770.34 KB 512x712 988.png)
I don't see any better place to post these, so here we go .... Enjoy.
(610.57 KB 512x688 3725.png)
(918.66 KB 720x920 1721.png)
(1.09 MB 560x808 1678.png)
S'more
(343.08 KB 512x512 a something.png)
>>137574 Dall-e has its own unique art style. And it knows that better to hind the hands than show them off, lol.
(226.72 KB 512x512 01.png)
(292.38 KB 512x512 02.png)
>137720 Else
here's a link to it: https://huggingface.co/WarriorMama777/OrangeMixs and to a pre-made colab notepad if you can't do AI on you PC: https://colab.research.google.com/github/camenduru/stable-diffusion-webui-colab/blob/main/abyss_orange_mix_2_webui_colab.ipynb
I wish I waz back
>>137722 i almost fucking pissed myself seeing the painting in the background. thank you for purple cat, it was technically a vtuber, and more like lavender cat, but its appreciated.
>>137724 love whenever AI adds extra random boobs in the corner of the picture. does this model have booru tags for characters?
>>137692 Nice — what model did you use for these?
>>137692 ayyy nice ones! What models are you using for these?
>>137730 Yea, most probably, didn't try that yet
>>137735 interested to see how this model handles those, if you do them please post
once again my PC has no idea what it is doing. DPM2 a Karras 40 steps. For some reason the last two images have had an attempt at text. tried "cat mouth" on the one before and the mouths were hell
>>137733 >>137734 Forthemasses model but it's really inconsistent. https://civitai.com/models/4673
>>137764 Does it work with USSBBW sizes?
>>137790 Nicely done! What model did you use for this?
>>137793 https://civitai.com/models/3950/art-and-eros-aeros-a-tribute-to-beauty this one
Has anyone had any luck making fat traps? Either photorealistic or drawn? Most models I've used don't handle androgyny well, and adding dicks into the mix just complicates things.
>>137804 >chicks with dicks
(368.24 KB 512x768 tmp7lvir4qz.png)
(441.06 KB 512x768 tmpq8bd3a_8.png)
(328.48 KB 512x768 tmpfkwh3fxw.png)
(451.94 KB 512x768 tmp_aqpy9lq.png)
(447.07 KB 512x768 tmp0e4fh18k.png)
(419.49 KB 512x768 tmprviloop4.png)
>>137726 In the colab version, how do you merge the model with the bigger girls model?
>>137853 Lmao, I recommend putting "teeth" and "open mouth" in negative prompts. for some reason it associates having teeth out with "nerd"-type prompts, which I noticed you're using, so be judicious about mixing tags with those
(308.14 KB 2065x1735 harold.jpg)
>>137853 >hide the pain >>137854 Maybe it's because anime girls are toothless or have the flesh teeth? >>137729 Hahah, thanks >>137574 >>137744 😭 >>137804 You gae
>>137837 See those !wget commands? They download stuff and put them in the directory specified after -O. So just make another wget command, put Bigger Girls download link in and put it in the same directory as all the others go, just name it bigger_girls.cptk or like that. After you boot up go to checkpoint merger, choose which model you want to be main(A) and which secondary(B) choose a ratio of secondary model in the final result (0-original A; 1-original B) and choose the checkpoint format based on what format was your main model, either safetensors or cptk. Don't touch the mixing method unless you do that with 3 models, 3rd starring stuff you DON'T want to be in final model.
IT'S OUT, I REPEAT, IT'S OUT! https://youtu.be/foeS4UBm-EU
>>137859 >>137861 >>137862 >>137863 Actually some pretty incredible stuff in here, color me surprised
>>137925 I liked it. Thanks for sharing, anon.
>>137946 Nah i'd say it's just amateurish
I imagined this chick I met a bar that looked kinda like this.....but more fuller....
>>137946 what model(s) are you using here? Nice works
>>137984 too horny to type
>>137946 Yea, dev probably meant this to be a thing, that's why he released both v2 and safe version with simplified prompting. Maybe try that one as well? I don't use it on it's own most of the time anyway, so I didn't see much difference aside from overall better shapes its mix with Anything 3 can do now
(500.73 KB 512x768 tmppr32w1pf.png)
(501.99 KB 512x768 tmp8m5if449.png)
(507.73 KB 512x768 tmpat4hy0_4.png)
(484.21 KB 512x768 tmpi_elncwo.png)
(471.39 KB 512x768 tmpgv5xo1n6.png)
(469.12 KB 512x768 tmp1iz2fkt8.png)
chub'd
>>137881 I tried adding the link and just copying the directory for the other commands, which made Bigger Girls appear in the selections under checkpoint merger, but I selected Abyss Orange as A and Bigger Girls as B and when I tried to merge them both boxes changed to ERROR in red regardless of the format I picked. What could have gone wrong? I notice a difference between the base four links and the last is that civitai doesn't end its link in either .safetensors or .ckpt if that makes a difference.
>>138021 huh, that's interesting. I mean, I tried merging v2 just this morning and this mf just crased my whole colab becuz it depleted all RAM, but I never got an error message with 1.2 and practically any other model. Here's how my code looks, check your code for errors maybe
>>138022 Oh I see what I did wrong. I had typed the site that hosted the download link, not the link to the direct file. Bigger Girls is apparently huge as a file so I get why yours crashed. How do you guys use img2img to improve your results? I get some neat stuff but with the occasional random error here or there that I'm sure is resolved with good img2img usage.
Stupid question, but which program does i need to make this kind of AI?
>>138034 Have a ripped PC with Nvidia graphics card? Wait for someone else, they'll tell ya how to work with client software, maybe. Got unlucky to have a weak machine or running it on AMD card? Go here: https://github.com/camenduru/stable-diffusion-webui-colab pick your model of choice and boot the thing up, our overlords at google give you a ripped fucking maching for 3 hours a day to do literally whatever. If you want stylized and anime images pick any version of Anything, Eimis, Elysium, Counterfeit, Certain, they can make kinda chubby gals on their own, but if you want to go broader, go to Civitai.com and take either BiggerGirls or ChubbyGirlsGolden, add them with a !wget command and either use on their own or merge with any model you want. It's a huge field, so you'll have to experiment on your own. If you want to make realistic fat gals images, you'll have to wait for someone who works with that. Adding a bunch of images I posted in deleted thread as a dummy guide to working with colab.
>>138034 >>138038 If anything goes south - kill the whole thing the most brutal way possible and check for everything to look somewhat simillar. First image is a way to kill everything, here >>138022 is about how wget section should look. All models come in either .cptk or .safetensors format, all of them require VAE files to operate properly, so check which models need what VAE's, usually they are listed on model page and come in their respectful colab file, if it's not done by retards that is
IMPORTANT! can someone make a stable diffusion guide for total beginners trying to make fatart! I want to start but im totally lost
>>138049 Well what models do you have?
>>137916 Link?
>>138038 how do i make those wget comands for bigger girls?
How do you guys get actually characters? I keep getting random OCs
>>138137 It's usually not enough to just punch in "[character], fat", you usually have to describe the character you want in detail. For example, image #1 of this post is just "gawr gura", and is completely unrecognizable as gura, but image #2 is "gawr gura, hololive, virtual youtuber, blunt bangs, grey hair, blue hair, blue hoodie, blue eyes, bare legs, shark girl, shark tail, shark hair ornament, shark hood, two side up, shark costume" with negative tags "blonde hair, purple eyes, grey hoodie, cat ears, animal ears, shorts, short shorts, skirt, miniskirt, thigh_strap, pants, leggings". Just go to gelbooru or danbooru, pull up an image for whatever character you want to replicate, then copy its tags. If your character isn't on a booru, or doesn't have many results, you'll have to either make an embed or get creative with prompting&inpainting. Then, once you are satisfied, you can start adding the tags you actually want. Images #3 and #4 have the gura prompt + some fat tags. And when you have a good image, if you want, can do some inpainting to fix details like hands, faces, clothes, anatomy, etc. A good candidate here would've been gura's shark tail or hands
(416.23 KB 512x768 tmphj1k_d86.png)
(370.97 KB 512x768 tmp1epid8p9.png)
(443.63 KB 512x768 tmpksi2gwm4.png)
(498.85 KB 512x768 tmpscqox3cu.png)
(437.60 KB 512x768 tmpv6cvfpf5.png)
(355.82 KB 512x768 tmpmu8nmd1l.png)
>>138049 I'm gonna copy and paste this from /vt/ >NAI Leak Speedrun https://rentry.org/nai-speedrun | https://rentry.org/sdg_FAQ | https://rentry.org/sdhypertextbook >/vtai/ Resources Holomem Prompt List: https://rentry.org/3y56t | VTuber Model Mixes: https://rentry.org/8nxtk | Embed/Thread Archive: https://mega.nz/folder/23oAxTLD#vNH9tPQkiP1KCp72d2qINQ >/hdg/ FAQ + Embeds, Hypernetworks, Models https://rentry.org/hdgfaq | https://rentry.org/hdglorarepo | https://rentry.org/hdgrecipes >Local Install Nvidia: https://rentry.org/voldy | https://github.com/AbdBarho/stable-diffusion-webui-docker CPU: https://rentry.org/cputard | https://rentry.org/webui-cpu AMD: Native: https://rentry.org/sd-nativeisekaitoo | Docker: https://rentry.org/sdamd | Onnx: https://rentry.org/ayymd-stable-diffustion-v1_4-guide >Cloud Hosted Install Colab: https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb | Paperspace: https://rentry.org/865dy Colab for Complete Retards: https://colab.research.google.com/drive/1STL60qfoY-iSlhRb9zFETRLTqhNbznRf >Subscription sites https://novelai.net/ | https://nijijourney.com/en/ | https://yodayo.com/ >Guides Prompting: https://rentry.org/hdgpromptassist Inpainting/Outpainting: https://rentry.org/drfar | https://rentry.org/inpainting-guide-SD Upscaling: https://rentry.org/sdupscale | https://rentry.org/hiresfixjan23 Animation: https://rentry.org/AnimAnon | https://rentry.org/AnimAnon-Deforum Wildcards: https://rentry.org/NAIwildcards Embeddings: https://rentry.org/simplified-embed-training Hypernetworks: https://rentry.org/HNSpeedrun LoRAs: https://rentry.org/lora_train | https://rentry.org/HDGLoRaIssues >Old News https://rentry.org/sdupdates3 >GPU Performance Data https://docs.getgrist.com/3mjouqRSdkBY/sdperformance >Upload your images https://aibooru.online/ https://www.pixiv.net/en/ >More Resources 4chanX Catbox userscript: https://rentry.org/hdgcb Essentially everything: https://rentry.org/sdgoldmine Tags: https://danbooru.donmai.us/wiki_pages/tag_groups /g/ wiki: https://wiki.installgentoo.com/wiki/Stable_Diffusion 5ch NovelAI wiki: https://seesaawiki.jp/nai_ch/ Models: https://civitai.com/ | https://huggingface.co/ | https://rentry.org/sdmodels More links: https://rentry.org/sdg-link
>>138094 Bigger girls Orange abyss v2
>>137515 I'm only a little late on responding to this, lmao. I couldn't give an answer until I booted up the program again because I normally don't boot it unless I know I'm gonna be busy, but I don't mind running it just to fetch the exact prompts again For the positive keywords I use cat girl, twin tails, long hair, messy bangs, (pink hair:1.1), (pink cat ears:1.1), black bows, purple eyes, pink cat tail and for the negatives I add blue eyes, black cat ears, white bows, purple bows, unibrow >>137688 I stopped using "nyatasha nyanners" when I realized that danbooru only had around 200+ results. I also didn't really have trouble with getting the ribbons to stay a certain color by just using the negatives. When I was originally setting it all up I just did some tests and added each negative color keyword until it stopped pushing out that color, and if it did push it out again once I started doing larger batches I would just change it when doing img2img
>>137688 Fuck, I forgot to ask what prompts and/or model mix you used for the Gray Wolfs here? Tossing them into the PNG Info didn't show many keywords
>>138438 You ever thought about opening an art account to rake in those delicious LIKES and VIEWS? This stuff blows away virtually everything else being put out by anyone, IMO. Mad cred to everyone else who is building the thread, as well.
>>138473 I mean, I drop my stuff to AIbooru from time to time, but come to think of it, why not make a brand new account for that somewhere on twitter or DA
>>138450 Was the second one done with the button gap LORA that's on civitai? I was wanting to experiment with it, but 99% of the prompts I have the AI generate normally involve swimsuits or latex so I didn't consider getting to it yet
>>138479 Holofat model? You've piqued my curiosity.
>>138495 Nah, I was just fucking around with a public ai site and got it by chance. I’m extremely new to ai, so I’ve been just experimenting with it all day. I’ve been using the available Abyss Orange model, and I’ve gotten some alright results.
i know asking for help is looked down upon here but I'm seriously lost. I've got Stable Diffusion AI, Anything v3 and Bigger Girls v2, but I can't figure out how to make it generate girls that look as good as the ones I see posted here. Is there certain prompts or a setting I need to change? The closest I got was feeding it an img2img and adding a specific character in the prompts, but even then the image still came out somewhat messy.
>>138506 ignore this, just saw the beginners guide posted a few posts ago
>>138507 >>138506 Unignore this, the "guide" doesn't really guide you, just gives you download links for things. I'm looking for specific prompts that will get me good fatties.
Finally tried fucking around a little with this on a webbased version (yodayo). I've gotten it to work reasonably ok but still struggling to get it to draw specific characters and not even obscure ones. Am I better off just installing the program and maybe trying img to img than continuing to tinker with a web based version?
>>138512 I dont know what to tell you, brother. Do you like huge breasts, or tiny ones?
fuckin around a bit, surprisingly fun stuff
(1.22 MB 720x824 00012-1770739494.png)
(992.88 KB 720x824 00021-1044165255.png)
(1.27 MB 720x824 00011-1834350101.png)
(1.24 MB 720x824 00013-1770739495.png)
(1.26 MB 720x824 00016-3046812896.png)
(1.30 MB 720x824 00017-3046812897.png)
More AI generated realisthicc fatties
(1.06 MB 720x824 00020-1044165254.png)
(1.01 MB 720x824 00022-2634975826.png)
(1.05 MB 720x824 00027-1868144893.png)
(1.10 MB 720x824 00023-2634975827.png)
(1.05 MB 720x824 00026-1868144892.png)
(1.23 MB 720x824 00029-3067658837.png)
(722.44 KB 536x664 00133-4038006264.png)
(812.36 KB 536x664 00103-1074169307.png)
(1.15 MB 720x824 00030-3067658838.png)
(1.14 MB 720x824 00061-2382088233.png)
(1.17 MB 720x824 00068-106931422.png)
(1.31 MB 720x824 00080-116162508.png)
(801.57 KB 536x664 00151-713898239.png)
(727.14 KB 536x664 00179-1610541998.png)
(766.76 KB 536x664 00166-4290023209.png)
(911.97 KB 2504x3124 have.jpg)
>>138566 Holy shit man XDD
>>138503 Scroll up to the /vt/ copy pasta and click "model mixes" and you'll find all the models in there including holofat.
>>138535 Whoa, these are great! What model/mix is this out of curiosity?
>>138603 Forthemasses model, just updated today BTW. https://civitai.com/models/4673/forthemasses
(534.03 KB 512x768 tmpa2brxny0.png)
(511.32 KB 512x768 tmp3kcrj5jq.png)
(551.24 KB 512x768 tmp1qibg_1d.png)
(526.08 KB 512x768 tmpy9qjtvca.png)
(515.97 KB 512x768 tmpuxq59c02.png)
fat foxgirl from my gacha game
>>138724 good stuff, we need more arknights fats.
Add voices to your pictures!: https://beta.elevenlabs.io/ AUDIO AI HAS BEEN RELEASED! Here’s an example of Patrick Bateman from a 4chan thread. https://vocaroo.com/1mcYA2LGqbJg
>>138793 anon while i do appreciate the funny drawn cat pictures it feels like a whiplash to see them under an ai fat art thread
>>137859 How did you get the colors so vibrant? It might just be me but the lines look clearer too. Been playing around with a few models and mixes like this one, and there's a number that seem to have a slight sepia haze I haven't been able to get rid of.
>>138891 Not the poster, but check that you're using the right VAE for your model. I know when I use the standard Stable Diffusion VAE it can have that effect. If it still happens, you could try out different VAEs, or try mixing your model a bit with one that has more vibrant colors. For a workaround fix, you could try adding "bright colors" or "high saturation" to your prompts.
>>138863 I don’t think you can fathom how proud I am of my kink and how god tier this innovation is
>>138951 You're a woman.
https://moritz.pm/posts/parameters This site has a big list of prompt keywords, I thought the lighting keywords were especially cool. I added "diffused lighting, rim lighting, volumetric lighting, hard shadows, subsurface scattering" on the last pic of this post and it did a pretty good job although its hit or miss (there were like a dozen other pics that gave the girl blue&black skin for some reason) >>138416 I merged anythingv4 & biggergirls and trained it a little more, mostly on pics of hips&asses My prompt was "masterpiece, best quality, highres, 1girl, grey wolf \(kemono friends\), animal ears, fur collar, black hair, yellow eyes, plaid pleated skirt,black thighhighs, cropped black suit jacket, grey necktie, sleeve cuffs, white gloves, fat, gigantic belly, midriff" and img2img alongside inpainting for the heterochromia & necktie+cleavage >>138479 Very nice details, I want to try and get something like this
>>138951 well congrats anon, now give me credit for the last pic because that art style is fucking mint.
(539.85 KB 512x768 tmpij_bhfkp.png)
(453.39 KB 512x768 tmp4g1lz36a.png)
(469.57 KB 512x768 tmp60azj80d.png)
(492.69 KB 512x768 tmpee9x6pdf.png)
(496.96 KB 512x768 tmp6i3nf6ii.png)
(534.85 KB 512x768 tmp0101zfgb.png)
>want to generate AI fatties >computer is probably too weak to handle running it locally >I have no idea how this shit works even if it somehow could >any and all information I can find on it (including the guides posted in this thread) requires you to already know a shit-ton about computers to use any of it apparently >try connecting to a colab instead >requires a google account >try to make a new one anonymously so I'm not tying this to my normal account >can't because it needs a phone number >disposable ones are always detected and not allowed as a result I apologize for being an absolute retard, but is there something I'm missing with either option, or am I just stuck paying the price for being dumb?
>>139255 Try making the spare google account on a phone itself? I've never get asked for a phone number that way.
>>139256 I could try that, yeah, issue is I don't exactly have a "burner" phone to do it with. Nor do I have the cash for one or ability to keep it from being known to others, being a penniless NEET. And I'm not about to tie my real phone to it for the same reason I won't use my real google account: Probably unfounded and pointless paranoia that people will find out I like fat chicks and mock me relentlessly for it. Stupid, I know, but hard fear for me to shake. I don't mean to fuck up the thread by the way, if there isn't an easy way to get around this (yet?), I'll just go back to silently enjoying the results without being able to add any of my own
>>139257 No shame in asking questions, especially if you're nice about it. It's not like it's your job to know these things (although i agree that you're worrying too much). it's a lot better to have a beefy computer for th e local install, but if you don't have a complete shit can, you can get away with running it on mediocre hardware (no less than 4gb of vram on your GPU) if you change some settings. You can also do cpu-only mode if your GPU is bad. It's incredibly slow but you can just leave it running in the background Nvidia users will have an easy installation, and AMD users are kinda shafted unless they're on linux. https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Dependencies Follow steps 1 and 2. Install Python and Git, and once you have git, you can clone (download) the program into whatever directory you want. Ignore step 3, since you're gonna want the fat girl model instead If that worked, you can proceed to setup, which is mostly automatic. Navigate into the downloaded folder and run webui-user.bat. It will take a while, and when it's done, close it if it doesn't close itself. After setup, download the bigger girls model: https://civitai.com/models/2664/bigger-girls-model and then put it in the models/stable-diffusion folder of your install. You should also change the startup parameters. Edit webui-user.bat and add the following, with quotes, to COMMANDLINE_ARGS= : "--medvram --opt-split-attention --no-half-vae" Sub --medvram with --lowvram if you have 4gb vram or less, works slower. --opt-split-attention helps reduce vram costs with a speed penalty. --no-half-vae makes increases vram costs but fixes an issue where images are solid black. The full list of arguments is here https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings After that you're pretty much set and only need to learn how to prompt, which is the easy part. Access SD by going to http://127.0.0.1:7860/ when the .bat is running. I don't know the specifics because I don't use windows or nvidia. I also don't use the SD colab so I got no advice there. But, hopefully this helps
>>139257 As long as your phone's an android, you should be able to make a burner account without a phone number. Hell, I made one for an androidtv device with a phone that's still on 6.0.
>>139041 Do you have any tips on training? Can you train an artist's style? I tried using dreambooth back in October and trained about 300 images, but it didn't work.
>>138724 >>139107 What model and prompts are you using, might I ask? I'm interested in generating my own arknights fats but don't know where to start
I know it's a long shot with how complex her design is, but having tried on my own for a bit and gotten nowhere with my rudimentary understanding of these programs, anyone willing to try and do the independent vtuber Juniper Actias?
>>139272 Also I forgot to mention that if cpu mode doesnt enable itself you can just replace --medvram with --use-cpu >>139298 I got a few Manually crop/resize & caption your training images, the autocrop is crap and autocaption tags every big girl regardless of size&shape as 'fat', making it really hard to train specific sizes or proportions, and really hard to train new concepts. Takes a while but the end result is better, you can use tags to generate, and you don't have to train as long. Just treat the file name as a string of tags in order of importance like "big_belly, fat, button_gap, office_lady, sitting_in_chair, brown_eyes" I also ran into this problem where including a specific character as part of the tags of an image made it a lot harder to generate them. i think it's because I didn't have enough training images on them. Whatever the problem is, it really sucks because now I can't generate patchouli knowledge without this horrible sketchy-cartoony style (see last image of this post) And while I haven't tried specific artists it looks like it's possible >>137863 >>138611 i found 60 steps per image on the default settings (with captioning turned on!) to be good. So, if you had 300 images, train for 18000 steps. Maybe 20000. You are also fine to not train the text encoder at all, since it's not needed if you've already captioned your own images. If you're on a free account the training takes a long time and you run the risk of being disconnected regardless of whether or not you're afk. So you should save checkpoints frequently or shell out $10 for premium Good luck
>>138535 now THESE are great
>>138793 I want to hear feederism audio
>>138535 I'm very curious what model or mix you used for this, simply because most of the models that use photorealistic sources don't do well with larger figures and these seem to be great.
>>139719 Not the poster, but this may be a mix of some model with "ForTheMasses", or just the latter on it's own. Didn't work with it, but stumbled upon it on civitai https://civitai.com/models/4673/forthemasses
Pardon my double post, but I have some questions relating to some of the models. >>139821 What is the other 70% you're mixing with Dreamshaper? I tried Bigger Girls v2 [F3C16C83DE] 70% with 30% DreamShaper [13DFC9921F], but it gave me [47f6f54edb] rather than your hash of [73610e09da]. Similarly, for>>137859 I tried 70% BiggerGirls-v1.2 Beta [FD3DDC7587] and 30% BaronMix [2C541F81CE] and got [4baaaa2f79] rather than [03079c9c55] My output is similar enough but I'm still curious.
>>139821 Might I recommend SwinIR_4x if you have trouble with Latent? You are correct to notice that Latent overcooks fat faces. I also notice that it tends to add extra fingers. Of the options besides Latent, SwinIR_4x is the one I ended up settling for. I used a prompt very similar to one of yours to run these tests.

Delete
Report