HimerosXL

(Making resources takes a lot of coffee, any donations are appreciated!)

Visit my Patreon/Gumroad/Ko-fi for exclusive access to all 90+ of my LoRAs! ▼▼▼▼▼

Members can get full access to my entire resource archive plus my Discord server for my steamiest content + direct feedback and previews! 💪

"Greetings Himeros, God of ?????? Desire..." 🏛️

Here is my first-ever merge release, HimerosXL. Designed to bring homoerotic content to SDXL's diverse and infinite possibilities!

Looking for additions to mix in? Here are my approved add-ons that work well with Himeros.

________ V2-06 ________

Version two is here! I rebuilt this from (basically) scratch. I have removed any and all DPO, most Turbo, and any other speedup/enhancements from this merge! This is to ensure the model is flexible with LoRAs and embeddings. You can use the Turbo, Lightning, DPO, or LCM LoRAs on top of this optionally, though I haven't tested them.

Even though I focused on photorealism with this, other styles are still possible, though V1 is more flexible. This is better at composition, prompt accuracy, and detail than V1.

This model is still focused on creating male portraiture, with ???? focus. It will give you ???? males without much prompting, SFW generations with clothing tags, as well as women, futas, and hetero content. I don't use it for those reasons so they aren't fully tested, but definitely possible.

For ????????, you can use more descriptors to try to get results you are looking for. "large ?????, small ?????, ????????, flaccid" etc. will work some of the time.

Currently, it has issues with realistic male asses, and multiple people in ?????? positions/interactions. (For better butts, you can add in LoRAs like Xbutt, Bussy, etc. but experiments including them here weren't very sucessful).

Because of the ??????? focus of this, you will get ??????? randomly, well, everywhere. Hands, on walls, floors, just be prepared for a sausage or two in your gens.

Future plans - I would also like more control over ??????? (size, cut/uncut, etc.) in the future. Better asses, as well as pose flexibility, and enhanced ????????, may be possible with further Pony inclusions in the future. I may also release a Pony-focused version of this that I have been working on.

A massive massive thank you to the authors of the ingredients of this! This merge contains -

• Jib Mix Realistic XL - v9.0

• epiCRealism XL - V5-Ultimate

• Acorn Is Boning XL - v1.0

• Virile XL - Alpha 1.0

• LEOSAM's HelloWorld XL - 5.0 GPT4V

• Xpenis - v2.13

• HimerosXL - V1-08 (see below for previous merges)

Recommendations:

• Sampler: Many samplers work well with this model. I have used many in my example images. I recommend trying XYZ Prompt with samplers to see their unique traits.

• Steps: 20-35 (dependent on sampler)

• CFG: 4-9

• Clip skip: 2

• Sizes: 832x1216, 896x1152, 1024x1024, 1152x896, 1216x832

• VAE: SDXL Vae fix

• Positive Prompts: Keep them as simple as "one man" or as complex as you wish. This model tends to understand both BLIP-style and Booru tags, I recommend mixing them both. Start your prompt with composition, then define specifics you want to see, and wrap it up with style. Sometimes full sentences work well.

• Negative Prompts: Keep them simple and to the point! Enough with the paragraphs of tokens, they typically make results worse.

• ADetailer: Absolutely recommended, use at 0.4 denoise on faces. Add in ????? Adetailer for enhancement as well, set your mask blur to 8. Replace your prompts for faces and ??????? with new prompts that removes ???????? from face prompt and body/face characteristics from ????? prompt.

________ V1-08 ________

I've been experimenting with this merge for awhile, and my members have been helping me test it before release. Please note, this is still in "beta", and like most XL stuff at the moment, isn't perfect. I will continue playing with merging in new content, and eventually probably train my own content to merge into it as well.

My goal with this merge is threefold - improve quality of ???????? and male physiques without needing extra LoRAs, remain stylistically flexible, and work fast. Therefore, it includes a recipe of many male-focused XL resources, as well as high quality base models that include LCM + Turbo. It also has a dash of DPO.

That being said, photorealism is currently what this model excels at. It works well, however, with Fooocus-style prompts to push styles such as anime, 3d, and art styles.

For ????????, you can use more descriptors to try to get results you are looking for. "large ?????, small ?????, ????????, flaccid" etc. will work some of the time.

You can expect this merge to give you great male portraits, which will mostly end up ???? even without prompting. That being said, it can do fully-clothed as well, and it can also produce women, futas, you name it. I have not fully explored what it is even capable of outside of males. It might even be better at hetero content than I realize, I'll leave that up to you! It can handle some ?????? positions/interactions, but expect deformities and struggles. Multiple people in a scene is also a nice feature it handles well.

I have tested some of my own LoRAs on this model, and the results are not amazing. I believe this is due to the Turbo aspect. I will attempt to correct this in future merges.

Because of the ??????? focus of this, you will get ??????? randomly, well, everywhere. Hands, on walls, floors, just be prepared for a sausage or two in your gens. Also, you'll get a lot of watermarks and URLs the more ?????? your prompts are. I'll work on this in the future merges.

A massive massive thank you to the authors of the ingredients of this! This merge contains -

• DarkBlueAphrodite - v2.0

• DreamShaper XL - Turbo DPMpp SDE

• Embrace The Suck (Alpha) - v0.71

• HugeAnimeCocks

• Tgirls XL

• RealVisXL V3.0 + Turbo

• Xpenis - v1.4

Recommendations:

• Sampler: DPM++ SDE Karras (Highly recommended but if it's giving you trouble, try Euler a)

• Steps: 7-10 (start with 8)

• CFG: 2-5 (start with 3)

• Clip skip: 1 (I haven't played with 2 but it might work?)

• Sizes: 832x1216, 896x1152, 1024x1024, 1152x896, 1216x832

• VAE: This should have a baked VAE. I still use the SDXL-Vae fix with it just to be safe.

• Positive Prompts: Keep them as simple as "one man" or as complex as you wish. This model tends to understand both BLIP-style and Booru tags, I recommend mixing them both. Start your prompt with composition, then define specifics you want to see, and wrap it up with style. Sometimes full sentences work well.

• Negative Prompts: Keep them simple and to the point! Enough with the paragraphs of tokens, they typically make results worse.

• Hires Fix: 8-12 steps, keep denoising strength between 0.15-0.3

• ADetailer: Absolutely recommended, use at 0.4 denoise on faces. Add in ????? Adetailer for enhancement as well, set your mask blur to 8.

I am not responsible for your use of this resource, which is not for commercial use and is for research purposes only.