Anymen v4 was the final update of the Anymen series . Please, check my other mixes here.
For those who want Anymen to live through their mix, read the section following the "Best Settings" section.
The Anymen mix is specifically designed to amplify male characteristics through the assembly of several base models. It allows the effortless production of top-notch male pics without having to solely depend on LoRAs, although you can still incorporate them if you prefer, as the mix is highly versatile.
Anymen mix gives an access to an array of semi-realistic, 2.5D and anime styles with the ability of effortlessly generate NSFW images.
Furthermore, it also offers a wide range of additional styles, thanks to the inclusion of BBFurrymix and Run diffusion FX.
My to go prompt :
Positive prompt : (1man), muscular, brown eyes, casual clothes, wearing a (white) tight short, big bulge, (intricate details, Masterpiece, high quality, best quality)
Negative prompt : UnrealisticDream, BadDream, FastNegativeV2, ((females, girls, girly, vagina, childish)),UnrealisticDream and BadDream / Fastnegative
Best settings
- Clip 1 will render 2.5D gens most of the time while Clip 2 will likely generate pics with a 2D digital painting style ;
- Euler A , Uni PC or DPM++ 2M SDE karras Sampler ;
- 20-35 sampling steps ;
- CFG scale from 5 to 9;
- Resolutions between 512x512 and 768x768 are recommended ;
- The AfterDetailer extension may be necessary to render clean faces and penis (Penis model for the AfterDetailer extension) ;
- The SD1.5 VAE is included but my favo one is Color101 ;
- Use emphasis for multiple characters, eg: "((duo, 2men)" or "((solo,1man))" or "((from below, low angle))";
- Here are my favo tags for NSFW pics "flaccid, penis, naked, saggy balls,..." ;
How to merge Anymen to a mix
Anymen is the culmination of several trials and errors, and countless hours dedicated to deeply understanding the merging processes.
The 4th update is the most stable and can be included in any mix, but my recommendations are the following :
- set Anymen at 0.45 - 0.5 as Checkpoint B ;
- if you plan to merge LORAs, merge them with higher values before adding Anymen in your mix ;
- as Anymen already comes with built-in penis LORAs, set a lower value if you are merging one.
Credits
Base models
V1-V2
Unstable Homoerotic Diffusion (UHD)
AnythingQingMix25D
Run diffusion FX 2.5D
V4
Indigo Furry
LORAs
BetterCocks V2
Anime penis
Turbo male
2 loras of mine
Description
additional merges and Color101 VAE baked in
FAQ
Comments (11)
FYI......too bad,.....V2 still will NOT load up or convert to diffuser in "InvokeAI"
Man... as mentioned in your previous comment, I can't help ya out on this. I never used InvokeAI and a discord member of the Unstable Diffusion who has an InvokeAI install had generated some pics using this model...
There is a small chance that the issue comes from your install.
@ewolf You can use a clip fix in A1111/Vladmandic to fix issue that occurs due to vae/clip being missing/f*cked when converting it to diffusers in Invoke.
Checkpoint itself will work though, but conversion is unavailable due to this problem.
As some one who created HomoDiffusion V1 to V3 & anime offshot I can tell you that merging in my model could cause some issues due to the fact that it has been heavily modified. The entire model & account dedicated to it was scraped due to this issue. If I decide to continue with V4 I'll have to start from scratch.
I think I found out why some people are having problems with the model. It looks like it's missing its CLIP. I'm not sure how you pulled that off, but I'm guessing at some point when you were merging models something happened and the CLIP was deleted or left behind instead of getting merged.
I discovered this as I've been testing dreambooth finetuning, and when I tried to use this model as a base, kohya_ss shit it's pants complaining about the CLIP. I loaded up the webui extension "stable-diffusion-webui-model-toolkit", and it said the UNET and Vae are fucked, and it doesn't even list CLIP as being present. (All models should have a UNET, VAE, and CLIP)
I'm guessing webui has some sort of contingency for when a model doesn't have CLIP included, which is why inference still works with webui, but not invoke. I have no idea why they would have a CLIP absence contingency, as this is the first time I've ever seen a model without a CLIP.
I'm guessing you need to re-trace your merging steps and try to figure out when something went wrong.
Thanks for letting me know :)
here is a fix : https://civitai.com/models/82527?modelVersionId=89225&commentId=155777&modal=commentThread&highlight=156141
@SeekerOfTheThicc the final update is up :) let me know if the model still has the same issue
It's great to see my half trained & half merged (HomoDiffusion) model survived in someone's else merge :D
Cannot wait to actually test it out when I'm gonna be back from holiday :3
Thanks 😁 it's such an honour
Here is my brand new mix https://civitai.com/models/91339/bb-male-mix-25d
@Deniryer the final update is up :)
@ewolf I'll review it once I'm done with training HomoDiffusion :)
I miss Deniryer - glad HD lives on in other ways.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.