Update AnimeXL and VelvetRender
image-gen
AnimeXL v7:
- The overall visual style has been significantly softened — less gloss or "glow".
- Some tags, such as "Mature female", now produce more accurate results.
- Minor improvements for prompts related to furry.
- Better handling of chibi and pixel-art styles.
- The model was further trained on more recent data (characters like Varesa from Genshin now work using only her tag).
- The likelihood of excessive brightness, desaturation, or overly orange/dark results has been greatly reduced.
- Reduced orange/sepia tint on many concepts (including yellowish bodies).
- Artist tags work better.
- Better results for more niche or abstract prompts, not just general ones like "1girl".
VelvetRender v3:
- Significantly increases realism, with crisper anatomy, more stable faces/hands, richer skin and hair details, and more natural lighting.