KouseiXL
Kousei (光星) is an innovative fusion aiming to synergize the distinct strengths of the HassakuXL (Pony) and HassakuXL (Illustrious) Style A variants. This model endeavors to provide a versatile anime art experience, leveraging the robust character generation and expressive capabilities of its Pony-based foundation, while infusing the bright, distinct anime aesthetic characteristic of the Illustrious influence.
The goal is to offer a model that balances NSFW potential with a polished, appealing visual style, making it easier to achieve high-quality anime artwork across a broader range of concepts. Kousei is a carefully crafted blend, with HassakuXL (Pony) forming the primary base, complemented by elements from HassakuXL (Illustrious) Style A, and utilizing the CLIP and VAE from the Illustrious version to help shape its unique output.
Using the model:
Important: Since this model utilizes the CLIP from HassakuXL (Illustrious) Style A, prompting strategies should primarily align with that model.
Recommended prompts: Use "masterpiece" in positive prompts.
Recommended negative prompts: "worst quality, bad quality".
The U-Net is a blend, but the CLIP strongly influences how prompts are interpreted.
Base models are trained with images aiming for minimal disruptive features. If "signature", "logo", "floating text", "speak bubbles" appear, add them to negative prompts.
Excluded metadata and franchise tags: Avoid using them. (e.g., "highres", "re:zero kara hajimeru isekai seikatsu").
Tag order can be flexible, but a common approach is: 1girl/1boy, character name, detailed description...
Width and height should be a multiple of 64.
Recommended resolutions for Kousei:
1024 x 1024
1152 x 896
896 x 1152
1216 x 832
832 x 1216
1344 x 768
768 x 1344
1536 x 640
640 x 1536
License info:
This model is a merge involving HassakuXL (Pony), which itself is a continuation of pony by PurpleSmartAI. Therefore, the most restrictive license applies:
This model is licensed under a modified Fair AI Public License 1.0-SD.
The following modifications have been added to Fair AI Public License:
You are not permitted to run inference of this model on websites or applications allowing any form of monetization (paid inference, faster tiers, etc.). This applies to any derivative models or model merges.
Description
Welcome to the first iteration of Kousei (光星)!
This v0.1 is an experimental fusion. The U-Net is built with HassakuXL (Pony) forming a substantial part of its foundation (around 70%), blended with HassakuXL (Illustrious) Style A (around 30%). Crucially, the CLIP and VAE are from the HassakuXL (Illustrious) Style A version, aiming to guide the aesthetic towards its bright and distinct anime style.
What to Expect / Goals:
Pony U-Net Influence: Potential for strong character generation and expressive capabilities from the Pony-based U-Net components.
Illustrious CLIP/VAE Dominance: Expect the overall style, prompt interpretation, and aesthetic to be heavily influenced by the Illustrious CLIP and VAE. This means a brighter, more distinctly "anime" look.
Balanced Output: The aim is to find a compelling synergy, offering a visually appealing and versatile anime model.
Early Observations & Considerations:
Prompting: Due to the Illustrious CLIP, prioritize prompting strategies recommended for HassakuXL (Illustrious) Style A. This includes using "masterpiece" in positives and "worst quality, bad quality" in negatives.
The specific U-Net ratio (Pony 0.7, Illustrious 0.3) is an initial attempt to balance the strengths of both parent models under the guidance of the Illustrious CLIP.
Experimentation is key. Observe how the Pony U-Net elements interact with the Illustrious-driven aesthetic.
Your feedback on this initial version is crucial! Let me know how Kousei performs, particularly regarding prompt responsiveness and stylistic output. Enjoy creating!