PixAI News: Exciting Updates!

New Models!✨

Introducing Hikari and Waterfront 0.9—our latest PixAI-trained foundational SDXL anime models, developed on our H100 cluster. (More and even better models are coming soon!)
Supports a wide range of art styles and character tags 🎨

Usage:

- Low-priority: ❌
- High-priority: ✅
- Turbo Mode: ✅

New Feature!✨

StyleMix: Effortlessly create with your favorite styles. Ideal for users new to style blending—combine with Hikari or Waterfront 0.9 for limitless creative possibilities!

Model Details:

Hikari:
An experimental model trained from SDXL 1.0 on a curated dataset of over 2 million high-quality anime images. Optimized for style tags that work seamlessly with StyleMix.

Waterfront 0.9:
A versatile model trained from SDXL 1.0 on a vast dataset of 10 million images (curated from a 50 million anime image pool).
- Supports over 2,000 character tags (7x more than Pony!)
- Supports over 5,000 art style tags, including combinations!
- For input formats, see the Waterfront 0.8 announcement; most remain unchanged, except for quality tags, which can now be customized or disabled as boosters.
Uses a dedicated sampler (sampler selection currently not available). We've also released Waterfront 0.9 Base for experimenting with sampler settings!

⚠️ Compatibility Note: LoRAs trained on other base models (e.g., Animagine, Pony) may not be compatible—please use with caution.

What will Coming Soon:

We’re releasing these two models' mvp version in advance for users to try them out with generation!
Meanwhile, a character tag reference sheet for these models is in progress. We also have a plan on benchmarks version for LoRA training, inpainting support, model comparisons, usage guides, and more!

Feedback and Issues

💡 Your feedback is crucial! Share your experiences and thoughts on the model-test channel.

Updated on: 26/03/2025