Here's the 1.80, introducing the StereoGen model. This model performs much better in inpainting, generating a realistic stereo view and minimizing the "halo effects" (one of the biggest 3D problems we have been tackling since the launch of Owl3D!). In this update, we also introduce a new Balanced V3 model, offering the same fast speed but higher quality. Please help us test it!
Features
- Introduced the StereoGen support for Nvidia GPU. StereoGen uses diffusion AI to generate much better stereo 3D visuals. Recommends: 16GB+ VRAM
- Introduced the Balanced V3 model based on distilled depth
Improvements
- Enabled xformers for Blackwell GPUs
- DepthGen - S supports very long scenes (one scene allows up to 5 minutes of 30fps video)
- Support Alipay for a perpetual license
- Improve UI performance when importing >100 images at the same time.
- Display payment URL in the purchase flow for copy-pasting.
Bug fixes
- Image degradation when blackbar detection + downscale was enabled
- Fixed VFR video detection on some platforms
- Improved A/V sync issue
- Fixed accidental enablement of low power mode when both Nvidia and Intel GPUs are available.
- Blackwell: Fusion mode is now properly supported
- Blackwell: Enable xformers for better performance
- macOS: Fixed ProRes output for >8K video
- macOS: Occasional Image warping error when depth intensity is too large
Changes
- Deprecated precision v1 model