Since last month’s roadmap update, you can now get vastly improved slow motion and FPS conversion with a new version of the Apollo AI model.
Better Apollo frame interpolation
We’ve released a completely new version of Apollo that produces significantly more natural results for difficult footage. This new release outperforms our older Apollo / Chronos AI models and decisively surpasses optical flow for footage with nonlinear motion.
Let’s compare slow motion using Optical Flow (left) vs Topaz Video AI’s new Apollo model (right). Notice the interpolation quality in areas of rapid motion, such as on the dog’s paws:
For this next video, follow the arm from the top left as it moves through the frame:
Finally, observe how natural the dancer’s hair and arms appear:
Introduced in v3.0, the first Apollo model specialized in interpolating frames for nonlinear motion and blurry footage. This new Apollo uses a completely new model architecture and improved dataset to achieve sharper and more natural results than the initial release. We’re very proud of these results and hope you find Apollo useful!
Use Apollo for up to 8x frame interpolation
Apollo always generates 8x frames from your input video and drops unneeded ones depending on your desired output. In general:
- Use Apollo when you require 8x or fewer frames
- Use Chronos Fast when you require more than 8x frames, or when reducing FPS
For example, use Apollo if you want to convert a video from 30 → 60 FPS and slow it by 400% since that requires 8x the input frames. If you’re not sure which model to use, queue up a preview for both Apollo and Chronos Fast, then toggle between the two versions to see which one you prefer.
We’re now quite satisfied with the quality of Topaz Video AI’s frame interpolation, so will now refocus on improving our general enhancement models.
We’ve also improved the stability, performance, and quality of the app over the past month:
- Slightly improve Motion Deblur quality and fix some edge cases
- Detect and re-download corrupted model files
- Temp files are deleted more consistently
- App now releases file when input file is released
- Performance benchmarking tool
- Upgrade to FFmpeg 6
- Fix various crashes and stability issues
- Improve tooltip and error messaging
- [Win] ProRes 422 HQ/LT output takes less disk space
- [Win] Nvidia H.265 output plays in Mac native apps
To see a full list of weekly changes over the past month, read the v3.1.9, v3.1.8, v3.1.7, and v3.1.6 release threads.
We’re actively working on a few priorities to improve the core value of Topaz Video AI:
- Improved quality enhancement model that gives more natural results with faces.
- Smoother in-app preview experience with better scrubbing / seeking and preview syncing.
- Further improve processing speed and better support running multiple processes.
Also on deck are more quality of life improvements like crash recovery, trim experience, and batch-downloading models. To improve general stability we’ve been actively investigating your bug reports and fixing appropriate issues in our weekly releases.
Thanks in advance for your feedback. We look forward to delivering another month of improvements to you!