Run this model free and uncensored on Tensor.art
Update - v2.0
Added Neotron -Welldone to merge
A little more flexible
Better penises
General NSFW probably improved but still very finnicky (tbh I recommend using SDXL/Pony/NoobAI for NSFW gens)
Images feel too dark actually, will try to fix that on the next update
I wasn't sure if I wanted to make an update, since there are way better gay NSFW models these days, but I actually still use SD1.5 for img2img so figured I would tinker with my own mix. Got some good results when merging with the Neotron model. Still not perfect when it comes to sex poses but honestly it probably never will. I wanted to focus on making the model more flexible with the men's appearances and poses while maintaining some realism and skin detail.
Helios Mix is my attempt at smashing some models together to hopefully create something interesting. I wanted to make a male model that focused on realistic faces but with some flexibility and creativity. It can do realism and 2.5D pretty well along with decent poses between two men, though not completely perfect. Been working on this for a few months but still needs some work so I appreciate all feedback.
Recommended VAE:
VAE is baked in but I still recommend this: vae-ft-mse-840000-ema-pruned.safetensors
My Settings:
Sampler: DPM++ SDE, DPM++ 2M Karras, DPM++ 2M SDE Karras, DPM++ SDE Karras
Steps: 20 - 30
CFG: 4 - 8
Clip Skip: 1
Hi-res Fix Settings:
Upscaler:
8x_NMKD-Superscale_150000_G (good for realistic details but kind of washes the image out)
R-ESRGAN 4x+ (good for stylized images that need smoother details)
4x-Ultrasharp
Steps: 15 - 20
Denoise Level: 0.35 - 0.55
Hi-res CFG: 4 - 5 (lower CFG seems to give less detail, which gives younger looking faces and vice versa)
Upscale by: 1.5 - 2
Issues/To-do list:
Fix the usual problem areas (eyes, hands, genitals, limbs, etc.)
Fix some of the weird artifacts (you might notice unwanted necklaces and earrings popping up)
Improve prompt understanding
More realism
Implement more of my own training
Models included in this merge are:
For the example images, I only use three negative embeds, which are:
If you want to support me, feel free to buy me a coffee.
Description
FAQ
Comments (6)
I'm surprised that this checkpoint isn't tons more popular. Its excellent.
Glad you like it. I'm guessing it's because people are moving on to SDXL now, especially with more gay models becoming available.
@Avrik I suppose that could be. But I have no room on my hard drive for SDXL models and LoRAs, so I will have to wait til I get my new laptop. I want more space and frankly more RAM before I do that. I will always use this SD 1.5 though because its nice and quick. I always use your model for most of my LoRA example pics. I have made some SDXL ones, but I have yet to test them or make example pics yet so it will be a spell before I upload them. Til then, this checkpoint will be my gold standard for example pics.
One of the several things I love about this model is that it doesn't seem to try to force itself into the facial structure being dictated by a lora. It seems to leave that totally up to the lora and I love that. With many other models, you can really see its signature stamped onto the face even if most of the facial structure comes from the Lore This one doesn't seem to do that, it seems very malleable to a lora's influence, which is why its my primary go to model for creating example images of my loras.
I haven't had the chance to test it out on too many loras so it's great to hear it works so well with them. Hopefully I'll have a chance to give it an update later, while keeping the flexibility in mind. AI stuff has been put on hold for me since I'm trying to save money. Was using a cloud GPU service for all of my merge testing and lora training.
@Avrik Take whatever time you need. I have to take breaks myself here and there due to long covid.