Actual goat 🙏

For some reasons your alpha values are 1e10 (approximated in bfloat16).

Snofs plus realism engine for editing from one image (man and woman for example) works pretty well

Just wanna share here,

9b base + snofs 1.2 (0.4) + unchained V2 (0.6) + turbo (1.0)
1.0 CFG, 6 steps, Euler

seems to give the best face consistency and quality for editing.

Any chance you can add footjobs to the mix? Great Lora!

1.2 is a huge improvement, keep it up dude! Much better at keeping face consistent in edit mode.

The new version is an improvement, but still unfortunately affects faces of character loras. Always makes them narrower, even at 1:1. The effect is stronger at other aspect ratios.

GOAT

Could someone recommend a good workflow for this?

Please make for 4B

Hello author of lora! The large model I use is Qwen-Rapid-AIO-NSFW-V19.safetensors. I simply use this large model, without adding any lora for image editing, the prompt text writes "The man enters this woman from behind, his penis is inserted into her vagina", the image effect generated in this way is very bad, although the large model itself has the nature of NSFW, but his details are still very poor, the woman's genitals are deformed, and the man's penis is also deformed, which is not a normal human sexual organ at all. Then I use the lora you developed, using the lora "snofs", I set the strength of Lora to between 0.5 and 1, or according to the original prompt, the generated picture is much more beautiful, the man's penis, the woman's pussy, are in line with the logic of reality, but, but there is a biggest problem that the penis is not inserted into the vagina, but inserted into the body area between the woman's vagina and anus, according to the descriptor to achieve insertion, but did not find a suitable position, directly inserted into the woman's body out of thin air, the labia vagina is always closed, just like a texture, can only be seen, but cannot be actually used. In addition to the penis inserted into the vagina, there is also the penis inserted into the anus, which is the same problem, the penis is not inserted exactly where I wrote in the description. How should I adjust, it has troubled me for a long time! The download link for the bottom model I used is: https://huggingface.co/Nouinhalegau/Qwen-Image-Edit-Rapid-AIO
The large model I use is the V19 version. The scheduler I am using is beta, the sampler is er_sde, the CFG is 1, and the noise is 0.85. How can I solve this problem? I uploaded two pictures, one man and one woman, and my descriptor said that their sex position was standing doggystyle, two people standing sideways facing the camera, facing the camera, and the penis was inserted into the vagina or anus.
Is there a problem with the AIO aggregated version of the large model I am using?

I use this lora as a photorealism pass for 'skeleton' SDXL semi-real images to nail the exact pose I want. The lora is extremely versatile and plays very nicely with aesthetic modifications with klein. Really really wonderful lora.

This is the best NSFW lora for both Klein and Qwen. I would love to see what this could cook with the more powerful Qwen 2512.

Thanks @Ashen3!

Does anyone have a good example or prompt for missionary position? I cant get anything near missionary. Other positions work fine. Great lora other than the issue im coming accross!

Sadly, this lora only makes tiny penises.

Great models/loras in the SNOFS series, highly appreciated. One comment regarding the Qwen line of the SNOFS loras: Qwen Image 2512 was released in the end of Dec. 2025 and is way superior to Qwen Image from before regarding anatomy and photorealistic rendering. I would say that it is much more reliable than Flux-2-Klein for complex scenes. Therefore I would like to ask if you may create a version 1.4 for Qwen Image 2512. Version 1.3 works in principle with Qwen Image 2512 but also causes some artifacts like vertical banding. Therefore such an update would be great. Thanks anyhow for your amazing work.

Z-image turbo version in plan? :D

Is there a trick to this, or am I doing something dumb? I can grab one of OP's example images for SNOFS 1.2 (klein 9b) and drag that image right into ComfyUI, exectute it, and get a wildly lower quality output. Really seems like I must have something borked, but I can't think what.

Hi, great model but 90% when i give a reference image it turns out with multiple or broken limbs. I tried lora at 1.0 and 0.6 but.. same. Any idea? THanks