Video Roadmap Update (Dec 2022)

Update: January 2023 Roadmap


In addition to stability, our team is currently focused on three core areas in Topaz Video AI:

  1. Faster export speeds. We’ve identified a significant opportunity to improve export performance and we’re in the process of implementing it. After we test and release it, we’ll continue to find new ways to improve performance like vendor-specific optimizations.
  2. Smoother in-app previewing. The existing in-app video player often displays frames slower than expected, especially on larger video formats. We’re working on improving in-app playback speed and overall preview experience.
  3. Better video quality models. In the short term, we’re testing a new motion blur correction model and improved Apollo frame interpolation variants. Afterwards, we’re working on better denoising and upscaling with improved parameter auto-detection, and also exploring options for face restoration.

These are the greatest current app bottlenecks for us to resolve before anything else. We will significantly improve export speeds and in-app previewing before considering other major features like pause/resume or additional encoder support.

Why v2 to v3

Topaz Video AI v3 was a complete rewrite from the ground up. While the previous version was a self-contained app, v3 instead provides proprietary FFmpeg filters and a user interface that spawns FFmpeg processes. Total app rewrites are not easy or fun, but there were several major advantages to this approach that caused us to make the leap:

  • Using FFmpeg means fewer video I/O issues in v3.0, better codec support, and better metadata/audio support. You can also extend or customize the FFmpeg commands to better fit your workflow.
  • Using separate UI and FFmpeg processes allows you to process in parallel, which better takes advantage of newer GPUs. The app can also recover from crashes and (later) resume paused or crashed videos.
  • v3.0 enables the development of new models and features that require multiple passes over the video. v2.6 treated video as a set of constant rate images, which makes new models like Stabilization impossible.

The disadvantage of a total rewrite, however, is that v3 doesn’t benefit from the years of bug fixes, stability improvements, and performance optimizations that went into v2. There are still some growing pains with the new app! While we’re improving Topaz Video AI every week, you can still install v2.6 alongside v3 if you prefer; the two major versions will not interfere with each other.

Please tell us what you think - feedback is a gift that spurs us to improve the product as fast as we can. We may not respond to everything, but we do read and consider every message. Thanks for your patience as we build a better app together.

29 Likes

Just my preference but I would rather see this as the priority as far as model development goes.

Thanks for the update.

9 Likes

from what suraj said in a previous post, if i understood well, the motion blur improvement will help actual and future models to get better. that would explain why it’s a priority over the denoise / upscaling.

8 Likes

VERY happy to hear that you guys are exploring options for face restoration. Face restoration is absolutely necessary for proper upscaling. Otherwise, you have sharp upscaled edges and a blurry face with no extra details. It doesn’t look great.

Thanks for the roadmap. I have more confidence in the team after reading this and I am excited for the future.

17 Likes

Really happy to see the roadmap post, keeping your customers in the loop is always a good thing!

6 Likes

Rewriting to FFmpeg was an excellent choice when looking into the distant future. It feels like the priority of the pause function has been pushed a little, but I believe that this function will be added someday. To the developers, as fellow developers, thank you for your hard work and dedication.

3 Likes

@eric thank you for that clear statement, and good luck to get it accomplished in time

1 Like

Topaz and its digital environment with other companies, a need for compatible formats .

Very well made your proposal for two lossless output layers for corrections in Adobe Final Cut, Black Magic, Blender, Nuke, Sony…

As the PSD format is a proprietary Adobe format, it may have implementation problems due to licensing difficulties, etc…

The DPX format as a 10-bit intermediate created by Kodak for use on telecine is my tool used as your plant in two layers, and for this I import twice into the timeline for final audio and video touch-ups.

In the current transitory situation of Topaz with many errors still unresolved, this DPX format available from the source in the aforementioned programs, implements a plugins called Cineon that allows very precise image adjustments at 10bits of range.

I lead a propagandist crusade in Topaz to implement this format that I have verified empirically up to 10 intraframe generations, almost without loss of quality, despite its age and it is free of rights or royalties.

This format is opened and allowed to work by PS- After Effect -Premiere- Audition- Media Encoder.and all other professional suites…

Color profiles and sound can be corrected, and once everything on your site is exported through Media Encoder or Apple Compressor to the final output format.

In times of such instability this is the process I use, losing a start generation to TIFF 16 bits. As it is not implemented in Topaz.DPX 10bits, let us continue to insist on its implementation as a temporary lifeline, since this format is implemented by the main commercial software, and it is really a very robust intermediate and a great help in restoring all image parameters…

Regards Eric Yang

Very good that Topaz has made a big announcement about the roadmap and the issues related to the rewrite.
I hope you will continue to update the roadmap once a month or so to ease the anxiety of users who have been waiting.

9 Likes

Faster export speeds Yes, please. I want to take full advantage of my water-cooled pc. I would also like to see the ability to change color spaces. Prores Raw to SRGB for example.

3 Likes

My problem is getting errors trying to Export after I get satisfactory results in Preview …
Here are some screen captures of problems … I’ve been trying to get past these problems for a week now … very frustrating …




How far in the roadmap is face recovery planned to be implemented? Basically any upscale of videos with faces is a lottery and not much can be done if there are faces upscaled with strong artefacts.

7 Likes

“exploring options for face restoration”

The biggest one for me. Have run into 3 unique times over the last month or so where Topaz was the better and often more perfect tool for a job but I couldn’t use it because I needed some level of face restore. Seeing as all my go-to options are free and open source for that – but tedious – I really do hope Topaz is able to put something in soon.

4 Likes

Will the ‘Comparison View’ be reintroduced into v3? I find the inability to compare and contrast the various models a great hindrance to workflow and aesthetic parameters.

5 Likes

I’ve got hope also, that AI face recovery function will be upgraded. I restored a lot of old videos from VHS on Topaz VEAI 2.6.4 and I was enjoyed. In all v 3.0 … are worse than 2.6.4 in this category.

1 Like



Dropbox

I once asked you to consider an option in the program so I could pause the conversion to a file when the computer needs to convert something else urgently, and the running Topaz Video AI would interfere with it. Will I see the change I was once promised on this forum? Thanks.

3 Likes

I like the new version. The improvement I would like to see is not strip/loose meta data. I use video AI to upscale and improve 360° video. The result is not 360° video as the meta data is not in the result. To work in Final Cut Pro the file needs to have 360° meta data injected back into it. Injecting the meta data back into many film clips seriously disrupts my workflow.

1 Like

read what the CEO wrote in his message, he answered your question :frowning:
I don’t understand why people write before reading changelog or announcement :frowning:

4 Likes

IMO, export speed may also be related to frame accurate seeking in ffmpeg as discussed here: svt-av1 (AV1 FFmpeg) significantly slower on PySceneDetect vs Equal Chunking · Issue #114 · Alkl58/NotEnoughAV1Encodes · GitHub

2 Likes