🌸 Rouwei from HaDeS v0.2
⚙️ What’s new in v0.2
✨ Improved overall image quality — sharper focus, cleaner details, and smoother rendering.
🌌 Enhanced backgrounds — better depth, color harmony, and structure.
🔵 Blue tone nearly removed — natural color balance restored.
🎨 Closer alignment to Illustrious from HaDeS aesthetic — softer light, cohesive shading, and finer tonal range.
⚡ More stable composition across complex prompts and long descriptions.
🌸 Rouwei from HaDeS v0.1
New RouWei-based finetune trained on 0.76M images.
🔮 Best prompt adherence among SDXL anime models.
🎨 50,000+ artists & styles with vibrant colors, smooth gradients without watermarks.
🖌️ Works with tags + natural text (hybrid captions).
⚡ Flexible: short or long prompts.
Sampling: Euler, Euler A - 20–28 steps · CFG 3–5
Quality tags: masterpiece, best quality / low quality, worst quality
👉 Separate artist/style tags with BREAK (or put them at the end).
❤️ Share your gens & support future training!
Description
⚙️ What’s new in v0.2
✨ Improved overall image quality — sharper focus, cleaner details, and smoother rendering.
🌌 Enhanced backgrounds — better depth, color harmony, and structure.
🔵 Blue tone nearly removed — natural color balance restored.
🎨 Closer alignment to Illustrious from HaDeS aesthetic — softer light, cohesive shading, and finer tonal range.
⚡ More stable composition across complex prompts and long descriptions.
FAQ
Comments (3)
Did v0.2 changed to epsilon or is still vpred?
it's still vpred
interesting finetune. Using rouwei as a base and building on top is a good idea. Will give this a go see how it compares to stock rouwei, which remains the king of diverse styles and difficult lighting. Does this have additional work done to it in addition to more recent training data?