artiwaifu-diffusion-1.0-GGUF

<h1 align="center"><strong style="font-size: 48px;">ArtiWaifu Diffusion 1.0</strong></h1>

The model is incredibly unpredictable and diverse. But this has a downside: the model is based on the first generation of stable-diffusion, is very prompt-sensitive and requires an explicit negative prompt.

---

<p align="center"> <img src="https://i.postimg.cc/RFN05PW0/1.png" alt="alt text" title="Cover" width="450"/> </p>

We have released the ArtiWaifu Diffusion V1.0 model, designed to generate aesthetically pleasing and faithfully restored anime-style illustrations. The AWA Diffusion is an iteration of the Stable Diffusion XL model, mastering over 6000 artistic styles and more than 4000 anime characters, generating images through trigger words. As a specialized image generation model for anime, it excels in producing high-quality anime images, especially in generating images with highly recognizable styles and characters while maintaining a consistently high-quality aesthetic expression.

Optimal Settings

• Sampler: Euler a, DPM++ 2M Karras

• Steps:

  • Euler a: <span style="color:red">50+</span> (min 13)

  • DPM++ 2M Karras: <span style="color:red">~35</span>

• CFG: <span style="color:red">4-9</span>

• Resolution: <span style="color:red">768x1192, 832x1216</span>

  • Width/Height: <span style="color:red">>=256px</span> (square images >=512px) are multiples of 32.

Prompt:

beautiful color, beautiful detailed, best quality, aesthetic, perfect body, detailed eyes, (Your Prompt)

optional*: by anime coloring, by cel shading

Negative Prompt:

watermark, worst quality, lowres, ugly, fused fingers, extra arms, missing arm, extra legs, missing leg, 3d, celluloid, sketch

The AWA Diffusion model is fine-tuned from Stable Diffusion XL, with a selected dataset of 1.5M high-quality anime images, covering a wide range of both popular and niche anime concepts up to April 15, 2024. AWA Diffusion employs our most advanced training strategies, enabling users to easily induce the model to generate images of specific characters or styles while maintaining high image quality and aesthetic expression.

Concise Guide to AWA Diffusion Prompting

Prompt Structure: (Style Tags) (Character Tags) (Scene & Action) (Aesthetic & Quality Tags)

1. Style Tags:

• Painting Style: View styles here
  • Examples: oil painting, watercolor, impasto
• Artistic Style: View artists here
  • Format: by artist_name (use with prefix 'by')
  • Examples: by asanagi
• Tips: More popular tags = better results; order and weight influence impact.
• Tips: Indicate the author with a distinct style (gishiki_(gshk), endou_okito, alterlesott, nakamura_regura)

2. Character Tags:

• View characters here
• Format: 1 character_name (series_annotation)
• Examples:
  • 1 lucy (cyberpunk)
  • 1 frieren
• Tips: Use exact trigger word, include series annotation if needed.

3. Scene & Action:

• Describe characters' poses, interactions, and environment.
• Examples:
  • sitting, arm support, smile
  • cowboy shot, gradient background

4. Aesthetic & Quality Tags:

• Quality: beautiful color, detailed, amazing quality (ranked highest to lowest)
• Aesthetics: perspective, lighting and shadow
• Rating: rating: general, rating: suggestive, rating: explicit

Examples:

• by yoneyama mai, 1 frieren, 1girl, solo, fantasy theme, smile, holding a magic wand, beautiful color, amazing quality.
• by nixeu, 1 lucy (cyberpunk), 1girl, solo, cowboy shot, gradient background, white cropped jacket, underneath bodysuit, shorts, thighhighs, hip vent, detailed, best quality.

Negative Prompts:

• Optional; avoid unwanted elements like signature, logo, or specific anatomical errors.
• Example: no glasses (for Tsuyu Asui)

Trigger Words & Tips:

• Typos affect results; be meticulous with spelling.
• Escape parentheses in tools using them for weighting.
• Preview tags on Danbooru for better understanding.

By following this concise guide, you can create effective prompts to generate desired images with AWA Diffusion.

Model Details

• Developed by: Euge
• Funded by: Neta.art
• Model type: Generative text-to-image model
• Finetuned from model: SDXL 1.0 Base
• License: Fair AI Public License 1.0-SD

<p align="center"> <img src="https://huggingface.co/raingart/artiwaifu-diffusion-1.0-GGUF/resolve/main/losses.png" alt="quantization losses" /> </p>

Support

buymeacoffee. Thank you for support!