Deep v1 (NoobAI V-Pred 1.0)

This is an experimental model. It makes good pictures but not every time. If you like it, please follow me.

Recommended Use

Base Model: NoobAI V-Pred 1.0

Weight: 1

Starter Prompts:

DBS, 1girl, 1boy, , faceless male, , masterpiece, best quality, newest

DBS, 2girls, , , masterpiece, best quality, newest

Look at my pictures for ideas and NoobAI page for model prompt guide. The character, and style, can change how good the picture is, so experiment.

Deep v1 (Pony)

This is an experimental model. It makes good pictures but not every time. If you like it, please follow me.

Recommended Use

Base Model: AutismMix Confetti

Weight: 1

Starter Prompt: score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, BREAK source_anime, rating_, 1girl, 1boy, , , , DBS, <lora:DBSPDv1:1>

Negative: grab

Look at my pictures for ideas. The base model, character, and style, can change how good the picture is.

Deep v2 (1.5)

This model is the result of countless days of work. If you like, please post your results and a review!

Recommended UseWeight: 1

Starter Prompt: <lora:DBSv2:1> 1girl, 1boy, hetero, , , , simple background

You can add more too. Look at training tags in Lora metadata to get ideas. This model does not work good with yuri but if you're patient it is possible. I will add more yuri in v>

Dimensions: 768x768, 512x768, 768x512

Each dimension setting works different, try them all out. Some gives certain positions more, some works better with other Loras.

This model works well with characters that don't have a strong style or pollution. It also works good with many styles.

Training InfoTrained on 113 hand picked pictures manually edited and cropped to reduce pollution. Trained on NAI for 4850 steps, this amount was chosen by evaluating models 50 steps at a time up to 6000 steps. Clip skip 2, x768.