A mix of random models that I did, I hope it'll be useful for you!
Recommended Settings:
Euler Ancestral or DPM++ 2M Karras
Steps: 25-50
Clip Skip: 2
Quality Tags: masterpiece, best quality, highres, absurdres, newest
Negatives: worst quality,bad quality,normal quality,bad anatomy,bad hands,error,watermark
The newest version does not dramatically change in terms of style, but rather is an attempt to improve the detail and anatomy!
(I really hope it improved compared to the last version...)
Description
Minor quality improvement and finally Megumin as the cover!
FAQ
Comments (5)
I wish there were more game-related characters! Still, I'm using them well for now! I'd also love to see some Wuthering Waves-related game characters!
Ah, thank you so much! I'll keep that in mind.
As a user with three years of experience in Stable Diffusion, I have tested dozens to nearly a hundred SDXL models. As of February 2026, considering my specific workflow and preferences, this has been the most "faithful," "consistent," and "adept" model for rendering the aesthetics of copyrighted characters—capturing everything from facial features to the distinct color palettes and styles of the original anime. I highly recommend it to anyone looking to reproduce IP characters with a high degree of fidelity.
While other models might certainly prevail when focusing on specific titles or characters, this model is second to none in its ability to render a wide variety of characters consistently well. The anime-style rendering, particularly the balance between flat shading and specular highlights, is superb. Anatomical accuracy (fingers and body) is average for an SDXL model, which is to say it's quite decent, and backgrounds are rendered with a respectable level of detail.
It is truly unfortunate that this model hasn't received the attention it deserves. However, I believe this is a significant example proving that other exceptional, underrated models are still out there.
Lastly, my sincere gratitude to the creator. Thank you. I look forward to your future work.
I typically generate images using ComfyUI with the following settings: Euler a/Karras, 25–35 steps, and a CFG scale of 4.0–6.5. My resolution is generally centered around 1024×1024, with a maximum of 1280×1280. Like other SDXL models, pushing the resolution beyond 1024 can occasionally lead to elongated limbs or the appearance of unintended subjects and objects, but this model's behavior in that regard is consistent with its peers. Since I utilize a workflow equivalent to Hires.fix for upscaling, I find 1024×1024 perfectly acceptable as a base resolution.
One thing to note is that skin textures can sometimes lean towards a 3D or realistic aesthetic. In such cases, adding quality tags like "(anime coloring:1.4), (official art, official style:1.2)" can be effective in guiding the output toward the desired look.
Thank you so much for the kind words! I'm grateful and happy to know that this model could be useful for you...