An experimental series of 2D checkpoints with no consistent theme. Generation settings may differ between versions and will be moved to their respective version descriptions if necessary. A VAE is already baked in.

Use the following generation settings:

For ComfyUI users, you MUST use the ModelSamplingSD3 node as shown below.

Positive Prompt: masterpiece, best quality

Negative Prompt: worst quality, low quality, jpeg artifacts, multiple views, comic, text, bad hands

Sampler: Euler A Comfy RF Normal, 24 steps

Resolution: 832x1216 (Portrait) / 1216x832 (Landscape) / 1024x1024 (Square)

CFG Scale: 5-6

Hires: 2x-AnimeSharpV4, Upscaled by 1.5x, Denoise Strength 0.15, 10 Steps

Shift: 6.0 †

† If you encounter any issues, ensure your WebUI supports the aforementioned sampler.

For reForge users, you may adjust your shift like so

Using adetailer is strongly recommended. Anzhc's YOLO face model was used for this gallery. Models used in this merge are

• Chenkin Noob XL v0.3 RF

• Kajinman Style LoCon

• minicop2001 Style LoCon

• frostedink Style LoCon

• HotVR Style LoCon

• roropull Style LoCon