/bbwai/ - bbwai

Stable Diffusion, Loras, Chatbots

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

no cookies?
E-mail
Password

(used to delete files and postings)
Misc

READ THE *RULES* BEFORE YOU POST!

[ / / ]
(1.88 MB 1500x1500 00008.png)
Good Generation Resources! Anonymous 03/04/2024 (Mon) 02:20:30 No. 21537
So, one thing that I've not seen on here is a good list of various models/loras that are good for generating things. The amount of them is mostly small, but finding them can be very difficult, because if they're not easy to know about in every case. So what I hope is that this thread can be for posting A. good models for generation of bbw, inflation, ect. and B. posting good LORAs for the same thing. The only time links need to exist, imo, is when the models can't be found with a google search or a civitai search. So, these are the ones I know about. Aurora Borealis: I don't know where this one came from, and I don't actually know where it's hosted at the moment; I only found out about it because people talked about it here. I've never gotten anything that good with it but other people probably have. BiggerGirlsModels: This one has a lot of artists in it but tbh I've never managed to ever get them to work well. The real version is good for a lot of people, and the 5.5 version is supposedly an improvement but I've just never really managed to get it to work. Hyperfusion: There's a LORA version and a model version. The LORA version, imo, is amazing and works with basically every model I've tried it on. The actual underlying model version I've, again, never gotten good results with. SymodelBiggerGirls: similar to the BiggerGirls models with a slightly different tagging structure, this is also the person behind the blueberry loras, I think? Speaking of, BlueberryInflation Lora and the Fluffyrock variety. Not bad, not great. Functional. There are several outdated models I don't know if I should put here, such as the BEplus lora. They are also functional. Again, the purpose of this thread is to consolodate all the various resources such as models and loras together so they can be easily found and searched. If, for example, you know of some that are good, or if you know of tricks that can be used for them, please put them here. Also, if you make your own, I suppose this would be a good place to put them too.
Anonymoose 03/04/2024 (Mon) 21:43:46 No. 21582
(85.13 KB 1024x1024 00022-4101032676.jpg)
I second the recommendation for the Hyperfusion LORA. Extremely useful for inflation work. However, does anyone have a good recommendation for a good checkpoint model? I've been using Dreamshaper 8 so far and it's good, but I would like some more tools for my toolbox.
Anonymous 03/04/2024 (Mon) 21:50:37 No. 21583
Perfect world v4 I like because it listens to prompts like an anime model but has a realistic rendering. Perfectdeliberatev40 is another I like for the same reason, way more flexible than super photo realistic models. Based66 is good for anime I've not tried other iterations tbh. Auroraborealis is good for simpler renderings and understands fat tags really well, combine with Lora's for best results. The thing I've noticed after using 20 or so models is you either get photo realistic models that are overly focused on one thing or flexible models that might mess up details but actually follow the prompts better. Also super realistic models can't handle Lora's nearly as well.
Anonymoose 03/04/2024 (Mon) 23:02:02 No. 21589
>>21583 Thank you, I'll try those out.
Anonymous 03/05/2024 (Tue) 18:06:40 No. 21635
Yea, the hyperfusion checkpoint takes a bit of getting used to, but once you find some decent artist tags/styles it out performs the LoRA on concept knowledge. The LoRA on the other hand just works since it uses the base models style.
Leaf 03/06/2024 (Wed) 08:29:20 No. 21653
VeldoraMix is a model that can get you some really nice fats in an anime style but it is skewed towards loli characters so make sure to take that into consideration when prompting with it.
Anonymous 03/08/2024 (Fri) 02:58:44 No. 21753
>>21582 Right now my favorite base model to use is the Obsidian comic style model. However, I've gotten good stuff with the aniverse models, and if you want anime stuff, aingdiffusion is a good one.
Anonymous 03/08/2024 (Fri) 15:32:36 No. 21766
>>21537 Ive personally wanted to stray away from Loras. Not because Loras dont work, but because I go for photorealism, and Loras remove life from gens, and make them samey. I think learning how to work a model alone is far more versatile, and produces better results, especially when used with Loras afterwords. Its very difficult to get good shapes on photorealistic models without any Loras. But ive found that tags like "Photoshopped" and "Photomanipulated" force models hand where they otherwise wouldnt go. Ive been interested in trying Prompt Alternating, because I think that could further larger proportions and much better shapes. Especially since generations tend to go either full perk or complete Lovecraftian blob. But I cant get it to work on my end. I suppose its only a feature in A1111 and not other UIs.
Anonymous 03/13/2024 (Wed) 05:40:50 No. 22002
The healing tool in gimp does a good job for getting rid of extra belly buttons and touching up other small deformations. The built in photo viewer in windows has something similar built in but can end up just leaving smudges, but works in a pinch.
Anonymoose 03/18/2024 (Mon) 02:54:22 No. 22238
>>22002 I have not used the healing tool, but I shall try that in the future. For belly buttons and hair color changes, I have been using the color picker tool and paint brush to cover up extra belly buttons, exporting that picture into a PNG, then using that PNG in img2img. Lets the computer do the hard work. This also helps a lot with changing a subject's hair color. As for GIMP, I have found that the Warp Transform tool is great for increasing the size of your subject's curves.
Anonymous 03/21/2024 (Thu) 04:07:53 No. 22328
(1.96 MB 1024x1536 unrefined.png)
(1.98 MB 1024x1536 refined.png)
>>21766 I figured out the refiner tab in a1111 also works on regular sd models. So you can gen an image with a non-photorealistic model and then refine/upscale it with a photorealistic model + latent hires fix. Almost gives pictures that uncanny dall-e realism. Also lets you use models instead of loras but so does controlnet.
fulforget 05/11/2024 (Sat) 11:59:15 No. 23937
>>21537 I use Arthemy Comics model (on Civitai). Good decent results - don't expect SSBBWs but for fluffy girls it gets the job done without Loras. I usually only need to add "(plump:1.2), weight gain" to my prompts.
Delete
Report