We’re excited to announce the release of the Vella-1.5! 🎉
This update brings significant performance enhancements, with especially notable improvements in automatic masking and the accurate representation of various lengths and fits. With Vella-1.5, you can now enjoy a more realistic and natural virtual try-on experience than ever before, taking a big leap forward compared to Vella-1.0.
Key Improvements
1. Overall Performance Boost
- Vella-1.5 shows consistent improvements across all supported categories compared to Vella-1.0. We’ve also enhanced outfit rendering from rear-view angles for a more seamless try-on experience. (More detailed updates will be shared soon via our technical blog 😊)
<Comparison with Vella-1.0>
2. Improved Fit Accuracy
- We addressed the previous limitation of inaccurate garment fit reflection.
- With enhanced masking preprocessing and training, Vella-1.5 delivers more natural fits that closely follow the intended shape and style of the garment image.
<Sample Results>
3. Better Length Representation
- We made significant improvements in handling length representation for bottoms (pants, skirts, dresses, etc.).
- The human model now generates more natural leg and lower-body lines, and reliably represents a variety of clothing lengths.
<Sample Results>
.jpeg%253FspaceId%253D12d3c8ec-852b-471d-949d-24da59b5d070%3Ftable%3Dblock%26id%3D25cc01da-7afe-802b-9340-fb276f1bc430%26cache%3Dv2&w=1920&q=75)
.jpeg%253FspaceId%253D12d3c8ec-852b-471d-949d-24da59b5d070%3Ftable%3Dblock%26id%3D25cc01da-7afe-8048-a7ac-d6f265ae1cb7%26cache%3Dv2&w=1920&q=75)
.jpeg%253FspaceId%253D12d3c8ec-852b-471d-949d-24da59b5d070%3Ftable%3Dblock%26id%3D25cc01da-7afe-80b2-a2f8-e29a1d35fe21%26cache%3Dv2&w=1920&q=75)
.jpeg%253FspaceId%253D12d3c8ec-852b-471d-949d-24da59b5d070%3Ftable%3Dblock%26id%3D25cc01da-7afe-805e-80ec-fae658a8d5fb%26cache%3Dv2&w=1920&q=75)
4. Enhanced Automatic Masking
- While we previously used an automatic masking model, it struggled with complex outfits or backgrounds.
- In this update, we applied a more advanced human parsing model for preprocessing.
- As a result, even without manual masking, the outputs are much more stable and consistent.
<Sample Results – Hairstyle Distortion Improvement>
<Sample Results – Arm Artifact Improvement>
.png%253FspaceId%253D12d3c8ec-852b-471d-949d-24da59b5d070%3Ftable%3Dblock%26id%3D25cc01da-7afe-8052-9020-c59e6ce83095%26cache%3Dv2&w=1920&q=75)
<Sample Results – Arm Masking Improvement>
Try It Now!
The updated Vella-1.5 model is available right now on Vella.
Check out the improvements yourself and don’t forget to share your feedback!
What’s Next
The Vella Team will continue improving model performance and will share more technical details on our blog soon.
Our mission is to make it easier and more convenient for everyone to create fashion content, and we’ll keep moving forward to achieve that.
💡 Try out the updated Vella today!
Share article