6chan/FeMix_HassakuXL

Femix HassakuXL

Overview

FeMix_HassakuXL is a series of Stable Diffusion XL models designed for generating anime-style images, employing a unique blend strategy to combine different aesthetic styles. The series ranges from highly stylized and experimental blends to stable, high-quality models suitable for general use. FeMix_HassakuXL merges two distinct models and concepts of each blend, to achieve various output depending on users needs.

Model Details

• Developed by: NeverbiasU
• Model type: Stable Diffusion XL blend
• Model prompt style: Booru-tags compatible
• License: Fair AI Public License 1.0-SD
• Blend Components: At v1 series, we only use Hassaku XL (Illustrious) v1.3 - Style B and Hassaku XL (Illustrious) v1.3 - Style A, while at v2 series we add one more Hassaku XL (Illustrious) v2 series model.
  1. Hassaku XL (Illustrious) v1.3 - Style B
  2. Hassaku XL (Illustrious) v1.3 - Style A
  3. Hassaku XL (Illustrious) v2
  4. Hassaku XL (Illustrious) v2.1
  5. Hassaku XL (Illustrious) v2.1fix
  6. Hassaku XL (Illustrious) v2.2
  7. Hassaku XL (Illustrious) v3.0

Recommended Settings

• Positive prompts:

masterpiece, best quality, detailed

• Negative prompts:

worst quality, bad quality, sketch, signature

• Sampling Method: DPM++ 2M Karras (Alternative: Euler A)
• CFG: 5-7
• Sampling Steps: 25-30

Supported Resolutions

1024 x 1024, 1152 x 896, 896 x 1152, 1216 x 832, 832 x 1216 (most recommended),
1344 x 768, 768 x 1344, 1536 x 640, 640 x 1536

General Recommendations

• Avoid overloading the model with lengthy prompts. Its versatility is also its weakness, as too much information can lead to unpredictable results.
• For high-quality faces: Put "nose" into the negative prompt for smoother faces, especially if Style A becomes too prominent.
• Excluded metadata and franchise tags, so please do not use them.

Things to Avoid

• Disruptive features: floating texts, logos, watermarks, speech bubbles, signatures
• If these issues occur, add "signature" to the negative prompt
• Excluded metadata and franchise tags (do not use them)

Version History

Version	Blend Composition
v1.0	Style A5B5 (50% A, 50% B)
v1.1	Style A7B3 (70% A, 30% B)
v1.2	Style A3B7 (30% A, 70% B)
v1.3	Style A4B6 (40% A, 60% B)
v1.4	Style A2B8 (20% A, 80% B)
v1.5	Style A1B9 (10% A, 90% B)
v1.6	Style A6B4 (60% A, 40% B)
v1.7	Style A9B1 (90% A, 10% B)
v1.8	Style A8B2 (80% A, 20% B)
v2.0	Style A5B5 + HassakuXL v2.0 CLIP/VAE
v2.1	Style A8B2 + HassakuXL v2.0 CLIP/VAE
v2.2	Style A9B1 + HassakuXL v2.0 CLIP/VAE
v2.3	Style A1B9 + HassakuXL v2.0 CLIP/VAE
v2.4	Style A2B8 + HassakuXL v2.0 CLIP/VAE
v2.5	Style A1B9 + HassakuXL v2.1 CLIP/VAE
v2.6	Style A2B8 + HassakuXL v2.1 CLIP/VAE
v2.7	Style A9B1 + HassakuXL v2.1 CLIP/VAE
v2.8	Style A8B2 + HassakuXL v2.1 CLIP/VAE
v2.9	Style A1B9 + HassakuXL v2.1fix CLIP/VAE
v2.10	Style A2B8 + HassakuXL v2.1fix CLIP/VAE [Unreleased]
v2.11	Style A9B1 + HassakuXL v2.1fix CLIP/VAE [Unreleased]
v2.12	Style A8B2 + HassakuXL v2.1fix CLIP/VAE [Unreleased]

License

This model is released under the Fair AI Public License 1.0-SD.

This model merges aspects from ANIMAGINE XL 3.0 (pre-V1) and Illustrious-XL & WAI-NSFW-illustrious-SDXL (V1 onwards).