⭐3wolfMondAI-LastSDG⭐

This remix of 3WolfMondAI-SDG was created to generate a realistic style without the need for realism or quality tags.

• Prompt recommendations
• Generation recommendations
• Sampler/Scheduler comparison
• RescaleCFG Grid
• Style examples
• Artist examples
• Dynamic Pose Grid
• Dance Grid
• Example Generations

⭐Model Description

• Developed by: Big Lasagna
• Funded by: /sdg/
• Also available on civitai

⭐Uses

A very special mix to try to create a model that generates (semi) realistic images without the need for realism or quality tags. It has a strong inbuilt style, and while it recognizes all db/e621 artist tags not all work well together with the inbuilt style (see examples below). Artist styles that are trained with a strong cartoony style will overwrite the realism and lead to toony characters in front of realistic backgrounds. If you want to try artist tags do weigh them mid-lowish, around 0.6, to try to combine the styles.

This model is in V-PREDICTION mode.

Prompts need to use danbooru/e621 tags, no Natural Language Support is added at this point. Use noobAI quality tags at your own discretion.

See below for various examples and comparison grids.

⭐Prompt Recommendations

All IllustriousXL/noobAIXL quality tags are recognized. These models work better with leaner prompts, so use as many tags as necessary and as little as possible for best results.

Standard Illustrious quality tags (e.g. masterpiece) have little effect so you can keep them or remove them.

Starter Positive prompt:

masterpiece, best quality, detailed background,

Starter Negative prompt:

worst quality, blurry background, blurry foreground,

⭐Improved Realistic Anthro Generation

A lot of danbooru and e621 tags come with a strong bias towards a more anime/anthro look of characters, even generic tags like "female" or "furry". To generate more realistic characters you should leave these and just prompt for a species and add some features that belong to one of the genders, usually the model will then generate the desired result. Depending on species there might still be a tendency towards a more standard furry look.

If you want increased fur detail, try adding some of these tags into the prompt:

(detailed fur:1.2), fluffy, shaggy fur,

If it still tries to pull your generations toward human features, if this happens you can add "(feature) human" to the neg prompt, e.g.

human hands, human lips,

⭐Generation Recommendations

The following settings are recommended starting points:

• Sampler: Euler A
• Scheduler: SGM Uniform
• Steps: 30
• CFG: 5-7

RescaleCFG of 0.5 is recommended, but depending on desired style you can also disable RescaleCFG.

Initial tests show that ancestral samplers work best but DPM can work in combination with the right scheduler. See examples below.

⭐More Information

⭐Sampler / Scheduler Comparison Grid

CFG 6 / RescaleCFG 0.5

⭐RescaleCFG Grid

⭐Style Examples Grid

CFG 6 / RescaleCFG 0.5

⭐Artist Tag Examples

CFG 6 / RescaleCFG 0.5

⭐Dynamic Pose Grid

⭐Dance Grid

Masterpiece, best quality, newest, No quality tags

⭐Example Generations (including metadata)

CFG 6 / RescaleCFG 0.5

Euler A / SGM Uniform	Euler A / Normal	DPM++ 3M SDE / Align your steps	DPM++ 2M SDE / SGM Uniform

Euler A / SGM Uniform	Euler A / SGM Uniform	Euler A / SGM Uniform	Euler A / SGM Uniform

⭐Uncommon Samplers Grid

CFG 6 / RescaleCFG 0.5