⚙️ Recommended Settings

🌸 Qwen Image 2.1

Tested with:
Qwen Image 2.1 int8_convrot/Noct Q v3 int8_convrot

• LoRA weight: 0.8

• CLIP strength: 0.8

• Steps: 25

• CFG Scale: 1.0

• Sampler: Euler

• Scheduler: Simple

🎨 Krea 2

Tested with:
CyberRealistic Krea 2/DaSiWa Krea2

• Steps: 10

• CFG Scale: 1.2

• Sampler: Euler

• Scheduler: Simple

• Denoise: 1.0

🐰 Anima

• LoRA weight: 0.8

• Steps: 40

• CFG Scale: 3.0

• Sampler: er_sde/normal

• Scheduler: Beta

✨ Illustrious

Tested with:
CyberIllustrious Semi and WAI

• LoRA weight: 0.8

• Steps: 40

• CFG Scale: 3.0–4.0

• Sampler: DPM++ 3M SDE GPU

• Scheduler: Karras

🐴 Pony

Recommended settings for the best results:

• Samplers: DPM 2 or DPM++ 2M

• Scheduler: Karras

• Steps: 20–40

• CFG Scale: 4–6

💠 SD 3.5

This version was trained on SD 3.5 Large. It has not been tested with other SD 3.5 checkpoints.

Recommended settings for the best results:

• Steps: 20–40

• CFG Scale: 4–7

• Scheduler: SGM Uniform

🔥 FLUX.1

This version was trained on FLUX.1 Dev, but it also works well with FLUX.1 Schnell.

The training dataset includes official character images and my own SDXL generations. I plan to continue improving the model in future updates to achieve more consistent and accurate results.

Recommended settings:

• Sampler: Euler

• Scheduler: Simple

• Steps: 20

These settings are recommended as a starting point. Results may vary depending on the checkpoint, prompt, seed, resolution, and LoRA weight.