A NSFW NoobAI XL Epsilon-pred 1.1 based model that produces photoreal images with much greater flexibility and scene composition prompt adherence than any other photo/real models I tested so far, including realistic pony models.

v0.3 Contrastes

Produces "realistic" images.

Added CyberIllustrious, IllustriousSat-IOR, 1girl printer 2.0, to the mix. Still NoobAI EPS 1.1 based.

v0.2 SummerBreeze

Produces "realistic" images.

Reworked the merging concept and reduced realism by ~20% to improve prompt adherence and image stability.

v0.1 PhotoReal

Produces photoreal images.

It is by no means perfect. As a child of a photoreal SDXL model, it will generate extra hands and bad eyes sometimes. Rarely, it may require ADetailer to fix faces for people in the distance.

And it is very naughty. :D

Same with my real Illustrious merge - ILove, I wanted to create a model that understands booru tags, knows (some) concepts provided by the amazing Noob-AI XL and produces photoreal images. In other words, I wanted a NSFW SDXL model that would provide greater adherence to complex prompts from scene composition perspective.

The complex multistep (~5 steps) MBW Dare & Ties merge included the following models:

NoobAI XL Epsilon-pred 1.1 - Base model for composition
Big Love XL - Base model for photo-realism
Cat Tower (NoobAI XL checkpoint) - Reinforcing composition

Enormous respect to the models creators.

Parameters

For realism: DPM++ (2M, SDE, etc.), Karras, Exponential

For smoother images: Euler A,

CFG: 4 - 7

Prompts

Keep it simple if possible. Excessive quality tags ruin composition adherence. Only put necessary tags in the Negative prompt.

I use:

Positive:

masterpiece, best quality

Negative:

worst quality, bad quality, text, watermark, bad hands, bad eyes, ugly,