IMPORTANT UPDATE: I used google colab to make this model, since some weeks ago, that's impossible to do, at least for free. I'm looking for some solutions, but untill that, I'm unable to keep updating this model, I'm sorry.
First of all, sorry if my english is a bit messed up as is not my first language.
This is my first test at merging models, with the intention of making a model capable of easily doing erotic anime images of big, hairy men. I probably did a lot of things wrong in the process, but im a bit happy with the result so far, and I hope you enjoy it. I will try to improve it bit by bit while I learn.
I find that this model can generate a wide angle of bodies without the use of plugins. If you don't want a muscular body, just type "muscular, muscles, bara" or similar in the negative prompt, it's quite flexible.
Using tags like "Anime Screencap", will give you an anime image (which was my original intention), otherwise it can do more realistic images too. Also you can use simple sentences for the gens.
The settings I use are:
Sampling method: I find DPM++2M SDE Karras. DDIM and Euler a the best options, but please experiment and share your results. I found in this last version that DPM++3M SDE Karras can work really well.
35 Steps
Size: 512 x 768
CFG Scale: 7-12 (but sometimes more if I use CFG scale fix)
High Res Fix:
12 Steps
0.4 Denoising Strength
Upscaler: 4x_NMKD-YandereNeoXL_200k or Anime 6b
1.5-2x Resize
I also use Tiled Vae and the default settings for Adetailer for face, and sometimes for hands.
Negative Prompt:
I found that "easynegative, negative_hand-neg, FastNegativeV2, Bad Artist Anime, woman, girl" works really well. But I'm not sure what would be the best on there. And I'm sure people more experimented than me could do better.
You will need a Vae for this, I personally use KL-f8-anime2, but blessed vae and Clear Vae work great too, you can download all of these here:
https://civarchive.com/models/118561?modelVersionId=131654
You can find the Dynamic Thresholding Extension and it's instructions here, for it's config you just have to use the settings it reccomends, it's really easy:
https://github.com/mcmonkeyprojects/sd-dynamic-thresholding/wiki/Usage-Tips
-Credits-
I merged mainly four models in different percentages:
Indigo Furry Mix v55
https://civarchive.com/models/34469/indigo-furry-mix
7th Anime v2_G
https://huggingface.co/syaimu/7th_Layer
Virile Animation
https://civarchive.com/models/101879/virile-animation
Endless Mix
https://civarchive.com/models/18375
V1.5 Additions:
Counterfeit:
https://civarchive.com/models/115569/counterfeit-v30-fp16cleaned
GSMaleto Fusion:
https://civarchive.com/models/126651/gsmaleto-fusion
Hassaku:
https://civarchive.com/models/2583/hassaku-hentai-model
V2 Additions:
Kizuki:
https://civarchive.com/models/22364/kizuki-anime-hentai-checkpoint
Criminal Art:
https://civarchive.com/models/136054/criminal-art-hentai-hardcore
+A bit more of 7th Anime v2_G and the 1.1 version.
Description
After getting really frustrated trying to mix and improve 2.0 various days, I decided to base this version on 1.5 again, using the same mixes of 2.0 in different percentages. Included models I used before or previous versions of this model with some additions.
I find it at least easier to use than 2.0 and the results are clean, and cleaner faces on group photos. I recommend to use Euler A or DDIM for easier results, but 2M++ SDE Karras and 3M++ SDE karras are good too.
I used the Kf8Anime and FtMse840000vae for the preview photos, with Adetailer and Tiled Vae. If there is a picture with more than 10 Cfg scale, I probably used the Dynamic Thresholding Extension with that one, I put a link in the description of the model if you are interested.
Also I want to merge some loras in the next version of the model, but It's really confusing, If someone knows how to use the Super Merger extension and can help me, I will be eternally grateful. It's hard to find any info online and I find the tutorial of the official page kinda cryptic.
FAQ
Comments (13)
Quickly becoming my go-to anime model! So so good, looking forward to testing 2.5
How to merge Lora to model:
1. Go to Supermerger tab
2. Under Supermerger tab there should be more tabs saying "Merge", "Lora", "Analysis" and more, you should pick the Lora tab.
3. Select what checkpoint you want to merge the Lora with under Checkpoint A and B (only choose 1 Checkpoint if you are planning to only merge 1 Checkpoint with a Lora).
4. There should be a list of Loras that you have downloaded onto your computer.
5. Choose a Lora and there should be something like: (Lora name):1, you can choose the weights by changing it like example: (Lora name):0.5.
6. Once you have chosen your checkpoint and the weights of what Lora you have chosen all you need to do is press "Merge To Checkpoint".
It is quite hard explaining it through text.
I am looking to make ugly bastards, Is this good at that?
It's quite flexible. Experiment a bit and share your results if you like. I'm still encountering new things lol.
this is still my favorite model for cock magic. I use it all the time, even occasionally like a tool with other models.
- making figures for other models to use w ControlNet openpose/depthmap/etc
- it's great to generate with then IMG2IMG upscale with something else with like VirileReality
- your implementation of tags, however you did it, not works v v v v well it also has taught me more about how tags work
Basically: behind every naughty cum splattered fella I make, theres stuff I learned or used directly from Hmen Mix.
It really is 12/10 man
Oh and one more thing: no other model ive ever found, including SDXL ones, does group stuff as well as HMenMix. Hell - ur prompts you've shared in the sample images of those group scenes alone have improved my cock magic immensely
Thank you so much
Tysm dude, im glad you enjoy it. I have to say, I myself don't understand tagging very well and I'm still discovering new things. This model is a work of mixing and mixing again a lot of models that I like. Surprisingly, I think the penis comes totally from indigo furry mix, when I tried to make human guys with it I discovered that It makes not only human penises really well, but a lot of positions and other things really well too, but with some furry traits. So my aim was to steer the posibilities of that model but to focus it into making dudes (And I tried to make it easy to use, but with every new version It's trickier I think :( ).
Indigo Furry mix uses e621 for the tags (https://e621.net/) and I haven't really test it, but maybe the tagging still works on this model. Sadly at least the onahole one, "penetrable sex toy" was an e621 tag that worked on Indigo models but I could not make it work on this one, but maybe a lot of them still work.
I wish that I could still improve it, maybe in a future when I have a proper income and I can buy either a good pc or Google colab pro.
everytime i try this model the images looks blurry, off colors and nonsense and idk why
hmm, the only thing that occurs to me is that maybe you are using the wrong lora or sampler? Just in case check the description, sorry for the bother.
I'm a total noob when it comes to this but whenever im generating images, the image during process looks so good but then when the image is done generating, extra limbs get added etc, I know I am doing something wrong and cant find a fix? I'm using suggested settings and VAE too, thanks!
Oh sorry for answering so late, I just saw that. Just in case you haven't found a solution yet, what it seems to happen is that you have denoise too high in high res fix, or that's what I think. More than 0.6 could make those things. I always have it at 0.45, try to gen like that and I hope it helps.
@FMen ah thank you for the response! I shall try it out, thanks :)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.