All my models are FREE, and all I ask is for your support by following, thumbing up and bookmarking! Thank you!

An all-in-one LORA for Z-Image (Base) that adds photographic cunnilingus capabilities in multiple configurable positions.

Trained with over 1,500 high quality images of cunnilingus photos.

Usage:

• Add the LORA at a recommended strength of 1. If using this with other LORAs, you can lower the strength as needed.

• You can use either "cunnilingus" or "oral sex" as trigger words, both works.

• Adding "vulva" to your prompt is helpful, as well as "licking her vulva" or "licking her vagina".

• Refer to the example prompt or gallery samples on how to best write up your prompts.

• It's best to add as much detail in both the characters and settings as possible; vague prompts will generally look bad.

• Z-Image is not the best at anatomy and multiple people interacting with each other, so if disfiguration appears, keep generating and you'll get optimal results!

Recommended Parameters:

• 30-100+ Steps (100 steps is ideal; however, 50 steps is a good balance of quality and speed)

• CFG 3-4

• res_multistep Sampler

• simple scheduler

• Shift 3

• 1024 x 1024 (square), 1024 x 1536 (portrait), 1536 x 1024 (landscape) resolutions

These are my recommended parameters only from limited testing. Feel free to experiment, as I'm sure there are better combinations out there.

Example Basic Prompt (Vague; Not Recommended):

A highly detailed photograph of a beautiful young woman, natural skin texture, realistic proportions. The woman is lying on her back with her legs spread apart. She has an expression of pleasure on her face, with her mouth open and eyes partially closed. A handsome man is positioned between her legs, performing cunnilingus on her exposed vulva, licking her vagina.

Example Detailed Prompt (Recommended):

A highly detailed photograph of a beautiful young woman, natural skin texture, realistic proportions. The woman has long, wavy brown hair and wears a white spaghetti-strap top with red polka dots and a pink plaid skirt that's been pulled up to her waist to reveal her exposed vulva. The woman is sitting on a transparent chair with her legs spread apart. She has an expression of pleasure on her face, with her mouth open and eyes partially closed. A handsome man is positioned between her legs, performing cunnilingus on her exposed vulva, licking her vagina with her hand on his head. The background is a modern living room and includes a beige sofa, a white table, potted plant and abstract colorful wall art in the background. Natural filmic color grading, gentle contrast, warm skin tones, soft emotional lighting, organic film grain, realistic cinematic beauty, grounded emotional mood, not glossy digital.

Avoid using "photorealistic" or "photorealism" in your prompts or else you'll get CGI image aesthetic types of output. Stick to words like "photograph" and "photo" only.

Seriously though, just get creative! You can change the pose to virtually anything—standing, sitting, lying, facesitting, on all fours with configurable arm and legs placements, various camera angles/shots and it should work.

If you need help writing up a specific cunnilingus pose, I recommend uploading an image to something like Joy Caption and modifying the prompt to your needs.

Does this work with Z-Image Turbo?

Yes, this LORA works on Z-Image Turbo, but with much reduced coherency. This LORA was trained on Z-Image Base and therefore is best used with Z-Image Base. So if speed isn't a concern, I highly recommend not using Turbo or Distilled versions. Setting the LORA strength to 1+ may be necessary if you are using Turbo or Distilled versions.

Why train on Z-Image Base rather than Z-Image Turbo?

I just don't find the quality and more importantly, versatility of Z-Image Turbo great. Z-Image Turbo tends to produce the very same looking outputs (including similar looking people), and has an infamous problem of seeding indifferences where even a completely different seed will produce nearly an identical output. Z-Image Turbo also has issues in particular with cunnilingus, and often adds a penis to a female. Z-Image Base was trained further, non-distilled and actually solves all of these issues. I know you just want to generate images as fast as possible or have a crummy GPU, but please take a look at Z-Image Base, as I find it absolutely awesome and currently my favorite model for photographs!

Because of all these problems, I likely won't dabble in training a version for Z-Image Turbo. It just isn't worth it for me.

Notes for V1.0:

• This LORA has NOT been tested on lesbian cunnilingus, as it was trained on heterosexual photos. However, if you have tested this LORA on lesbian cunnilingus, could you please report your findings in the comments below for myself and others?

• This LORA should NOT be used to generate anime or cartoon images; however, 2.5D at lower LORA strength can still work.