AnyHentai is a merge of multiple models
for sampling methods, use Euler a (best), DDIM (second best), or DPM++ DM Karras
DOWNLOAD/INSTALLATION
Step 1: download SAFETENSORS and VAE files.
Step 2: put the SAFETENSORS file under "stable-diffusion-webui\models\Stable-diffusion"
Step 3: put the VAE file under "stable-diffusion-webui\models\VAE"
Step 4: Done! enjoy the model
USAGE
-use a minimal negative prompt for best results
-use Euler A and 20-25 steps for best results
-use danbooru tags
-I used a clip skip of 2 (optional)
-I also used the upscaler Latent (nearest-exact) with a highres step of 20 and denoise of 0.5 to improve image quality and detail (optional)
DO NOT USE A DENOISE STRENGTH BELOW 4.5 FOR BEST RESULTS
EXAMPLE PROMPT
((Masterpiece)), (best quality), (1girl), red hair, beautiful red eyes, medium breasts, classroom, black glasses, school uniform
VAE
The VAE is not required. you can use any VAE you like. I have found that VAE makes more vibrant and crisp images, but this is just my testing
for the VAE I used the same one as grapefruit
I will try to improve and update this model by adding other images, although I'm not too familiar with SD and training models, so I will most likely stick to only merging models.
_______
EXTRA INFO
-Loras have not been tested yet, but they should most likely work
-use the upscaler Latent (nearest-exact) for best results
- I will try to update the model at least once per week
-Image generation on the A1111 WebUI normally took around ~10 seconds using a 3080 TI with 12 gigabytes of RAM
CREDIT TO THE PEOPLE WHO MADE THE MODELS ORANGEMIX, HENTAI DIFFUSION, ANYTHING V3, ANYTHING V5, DREAMSHAPER, KENSHI, COCOA, CORNEO'S 7TH HEAVEN MIX, PROTOGEN V2.2 (ANIME), KOTOSMIX, COUNTERFEIT, MIXPROV3 AND GRAPEFRUIT
Description
MUCH BETTER ANATOMY
merged new models (COCOA, CORNEO'S 7TH HEAVEN MIX, and PROTOGEN V2.2 (ANIME))
much better overall
FAQ
Comments (9)
Omg man, you're a star!
Thank you very much for the positive feedback! It helps alot!
Does it have to be 7 Gb+? This is the most heavy mix I know.
I can find a way to decrease the size for 1.7! thank you very much for your feedback.
briefly: according what are you doing there is a question coming Why even ur model needs if any4.5 already exists
i understand that you try hard to be useful but please reconsider what u do.
First - FYI: than more u will keep merge various models than the final ckpnt will be lose an knowledge from previous, that means regression of project.
Second - according what u merge(given your description from) almost each ckpnt include same model in some % of weight(Any3.0), that means you are literally spin in circle(merging) over and over again a single model and that means eventually you get Any3.0 more than ~75-85% weigh from final model.
of course i'm not your mom to make what you do, but i'd try merge by weight blocks to improve details rather to merge entire models (because eventually it'll getting an useless garbage)
@HTKE I don't know much about merging since I am very new to AI and SD, but I will take your advice into account for when I make 1.7. I didn't know I could do this! Thank you very much
how do i turn SafeTensor file into ckpt???
Some GUI's like NMKD Stable Diffusion can convert SafeTensor files into Checkpoints. Other GUI's like InvokeAI can use the SafeTensor files.