Agreed, the estimates are useless for now. I rely on the temp file to get an idea of what’s done vs. remaining.
I am transferring HDV (1440x1080) interlaced tapes to 4K. I have used both Proteus and the miniDV basic (DioneDV). I cannot decide which one I like best. Anyone else done this format and what did you like? Not concerned about speed just best quality. Thank you
2nd video, similar results. Good news is it’s stable and results are good. 0.2fps on Mac Studio M4 Max 36GB with macOS Tahoe 26.3 beta 3
- Video: h264 (Main) (avc1 / 0x31637661), yuv420p(tv, smpte170m/bt470bg/bt709, progressive), 640x476 [SAR 423:320 DAR 423:238], 484 kb/s, 25 fps, 25 tbr, 90k tbn (default)
- frame=30956 fps=0.2 q=-0.0 Lsize= 538593KiB time=00:20:38.11 bitrate=3563.6kbits/s speed=0.00818x elapsed=42:04:04.62
cok güzzel 
1 Like
davinci resolvede olusuturdugum lut ıle render ettim
Something else: “Proteus” or “Iris MQ” with add Grain “Gaussian” at density approx 40%-50% AND as a second model choose “Gaia” delivers an awesome grain result! Unfortunately, the rendering time increases enormously and Gaia has other effects, it fits not for every footage.
This is for input if Topaz personal is reading along: It would be nice if Grain could be used independently without a model selection. Please take a look at what I mean with, for example, Proteus Natural with add Grain Gaussian + second model Gaia. I simply see potential here for creating an independent add Grain-Gaussian-Model, that can deliver same results.
2 Likes
Getting errors trying to use the Proteus Natural model, I’m just upscaling a short 1080p video to 4K, tried using a H.265 and H.264 version. Running on M1 Max MacBook. Checked the error logs:
2026-02-04 19-50-30.933 Thread: 0x2054e2c40 Info File Loaded successfully: 0 /Users/dominic/Desktop/Upscale Tests/Show Road.mov
2026-02-04 19-50-30.933 Thread: 0x2054e2c40 Info File Loaded successfully: 0 /Users/dominic/Desktop/Upscale Tests/Show Road.mov
2026-02-04 19-50-30.937 Thread: 0x2054e2c40 Info File Loaded successfully: 1 /Users/dominic/Desktop/Upscale Tests/Show Road.mov
2026-02-04 19-50-30.937 Thread: 0x2054e2c40 Info File Loaded successfully: 1 /Users/dominic/Desktop/Upscale Tests/Show Road.mov
Format mov,mp4,m4a,3gp,3g2,mj2 detected only with low score of 1, misdetection possible!
moov atom not found
2026-02-04 19-50-34.466 Thread: 0x2054e2c40 Info FF Process Output: 43
2026-02-04 19:50:34 0x1702d7000 INFO: PARAM DEVICE: -3
2026-02-04 19-50-34.489 Thread: 0x2054e2c40 Info FF Process Output: 43
2026-02-04 19:50:34 0x1702d7000 INFO: PARAM DEVICE: -3
2026-02-04 19-50-34.513 Thread: 0x2054e2c40 Info FF Process Output: 43
2026-02-04 19:50:34 0x1702d7000 INFO: TOTAL INSTANCES: 3
2026-02-04 19:50:34 0x1702d7000 INFO: Init backend runner 3
2026
2026-02-04 19-50-34.514 Thread: 0x2054e2c40 Info FF Process Output: 43 -02-04 19:50:34 0x1702d7000 INFO: Moonlight: set batch config 8 0 0 2
2026-02-04 19:50:34 0x1702d7000 INFO: [Pipeline]Start layer: model
2026-02-04 19:50:34 0x1702d7000 INFO: [Pipeline]Connecting model to blend
2026-02-04 19:50:34 0x1702d7000 INFO: [Pipeline]Connecting blend to post
2026-02-04 19:50:34 0x1702d7000 INFO: Auth Check Watermark: 0 0 0 2026-01-01 2026-09-29
2026-02-04 19:50:34 0x1702d7000 INFO: Update status every 4 frames
2026-02-04 19:50:34 0x1702d7000 INFO: [Pipeline]Connecting post to status
2026-02-04 19:50:34 0x1702d7000 INFO: [Pipeline]Output layer: status
2026-02-04 19:50:34 0x1702d7000 INFO: Video Processor setup successfully for model pnat-1
2026-02-04 19:50:34 0x1702d7000 INFO: With input dimension (width x height) 1920 x 1080
2026-02-04 19:50:34 0x1702d7000 INFO: With output dimension (width x height) 3840 x 2160
2026-02-04 19-50-34.515 Thread: 0x2054e2c40 Info FF Process Output: 43 e[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;226mdeprecated pixel format used, make sure you did set range correctly
e[0me[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0me[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x850fc8000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0m
2026-02-04 19-50-34.515 Thread: 0x2054e2c40 Info FF Process Output: 43 e[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x850fd8000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0me[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x850fe8000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0m
2026-02-04 19-50-34.515 Thread: 0x2054e2c40 Info FF Process Output: 43 e[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x8513c8000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0me[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x8513d8000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0m
2026-02-04 19-50-34.515 Thread: 0x2054e2c40 Info FF Process Output: 43 e[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x8513e8000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0me[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x850504000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0m
2026-02-04 19-50-34.515 Thread: 0x2054e2c40 Info FF Process Output: 43 e[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x850514000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0me[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0m
2026-02-04 19-50-34.515 Thread: 0x2054e2c40 Info FF Process Output: 43 e[48;5;0me[38;5;153m[swscaler @ 0x850524000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0me[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x850534000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0m
2026-02-04 19-50-34.515 Thread: 0x2054e2c40 Info FF Process Output: 43 e[48;5;0me[38;5;153m[swscaler @ 0x8557f0000] e[0me[48;5;0me[38;5;153m[swscaler @ 0x850544000] e[0me[48;5;0me[38;5;226mNo accelerated colorspace conversion found from yuv420p to rgb48le.
e[0m
2026-02-04 19-50-34.519 Thread: 0x2054e2c40 Info FF Process Output: 43
2026-02-04 19:50:34 0x1702d7000 INFO: ---StandardTilingStrategy::StandardTilingStrategy W: 672 H: 384 C: 3 R: 3 X: 624 Y: 348
2026-02-04 19:50:34 0x1702d7000 INFO: ---StandardTilingStrategy::StandardTilingStrategy W: 1344 H: 768 C: 3 R: 3 X: 1248 Y: 696
2026-02-04 19-50-56.583 Thread: 0x2054e2c40 Info FF Process Output: 43 2026-02-04 19:50:56.583 ffmpeg[90098:26850172] Output Err: Unable to compute the prediction using ML Program. It can be an invalid input data or broken/unsupported model.
2026-02-04 19:50:56.583 ffmpeg[90098:26850172] Deleting model cache directory: /Applications/Topaz Video.app/Contents/MacOS/../Resources/models/coreMLCache//pnat-v1-fp32-384x672-2x-mlp
2026-02-04 19-50-56.587 Thread: 0x2054e2c40 Info FF Process Output: 43
2026-02-04 19:50:56 0x17182b000 CRITICAL: Unable to run model with index 0 it had error:
2026-02-04 19-50-56.587 Thread: 0x2054e2c40 Info FF Process Output: 43
2026-02-04 19:50:56 0x17182b000 CRITICAL: Caught exception in tile processing thread and stopped it msg: std::exception1libc++abi: terminating due to uncaught exception of type std::exception: std::exception
2026-02-04 19-50-59.739 Thread: 0x2054e2c40 Info process exited error occurred: 43 1
2026-02-04 19-50-59.739 Thread: 0x2054e2c40 Info process exited: 43 6 1
2026-02-04 19-50-59.739 Thread: 0x2054e2c40 Info Going to delete path: "/Users/dominic/Desktop/Upscale Tests/Show Road_395752552.mp4"
2026-02-04 19-50-59.739 Thread: 0x2054e2c40 Info Close and delete path "/Users/dominic/Desktop/Upscale Tests/Show Road_395752552.mp4"
2026-02-04 19-50-59.739 Thread: 0x2054e2c40 Info Going to delete path: "/Users/dominic/Desktop/Upscale Tests/Show Road_pnat1.mp4"
2026-02-04 19-50-59.739 Thread: 0x2054e2c40 Info Close and delete path "/Users/dominic/Desktop/Upscale Tests/Show Road_pnat1.mp4"
2026-02-04 19-50-59.741 Thread: 0x2054e2c40 Info ~TProcess(): destroyed
2026-02-04 19-50-59.755 Thread: 0x2054e2c40 Warning qrc:/videomanager/TVideoManagerPane.qml:1245:29: QML WIconImage: Detected anchors on an item that is managed by a layout. This is undefined behavior; use Layout.alignment instead.
2026-02-04 19-51-05.126 Thread: 0x2054e2c40 Info Logging files: 15
2026-02-04 19-51-05.180 Thread: 0x2054e2c40 Warning Showing in graphical shell: "/Users/dominic/Library/Application Support/Topaz Labs LLC/Topaz Video/logs/logsForSupport.zip"
Can you send the full log to the support team at help@topazlabs.com so we can assist further?
So Chat GPT is genius for such things…
- moov atom not found
Format mov,mp4… detected only with low score of 1
moov atom not found
This means FFmpeg can’t properly read the MOV file structure. The moov atom contains metadata (duration, frame index, etc).
2. CoreML model crash (
this is the fatal error)
Unable to compute the prediction using ML Program.
invalid input data or broken/unsupported model
Deleting model cache directory:
pnat-v1-fp32-384x672-2x-mlp
CRITICAL: Unable to run model with index 0
Conclusion
The **PNAT v1 model failed to execute on Apple CoreML, then crashed the tile-processing thread. This is not a normal FFmpeg error — it’s the Topaz ML backend failing.
Why this usually happens on macOS
One (or more) of these is true:
A. Corrupted CoreML model cache (very common)
Topaz even hints at this by deleting the cache automatically — but sometimes that’s not enough.
B. Model + input mismatch
PNAT is very picky about:
- Bit depth
- Colorspace
- Pixel format
- Frame dimensions
and there is much more including steps to try to fix, just copy/paste the log into ChatGPT
Quit Topaz completely, then delete:
~/Library/Application Support/Topaz Labs LLC/Topaz Video/models/
~/Library/Application Support/Topaz Labs LLC/Topaz Video/coreMLCache/
Reopen Topaz and let it re-download.
2 Likes
>This issue is being worked on…
So you guys are aware of this and still bill people? If you remove an option making the program unusable, shouldn’t you offer that time (over 5 months in this case) for free or some other compensation? Or is this the reason why you changed to a monthly subscription to rip people off even when they can’t use the software?
>Can you send the support team your logs
Why? To get another copy pasted reply? First telling me to “turn max memory to 80%” or something? Then telling me that I shouldn’t complain? Then telling me that you are aware of the problem and don’t give a shit? Your support hasn’t been helpful in any problem that I’ve ever sent them.
You guys are aware of this problem. You guys are aware that it worked perfectly before 1.0.3. NOW FIX IT!
1 Like
How is frame interpolation still borked a year later?
The interpolated output “may” initially show constant 60fps, but when doing a further compression ensuring to use the timecodes from the “constant 60fps” video it tells a different story in the output..
Sure, its close enough where you could disregard the correct timecodes, force constant 60fps and be unlikely to notice any obvious A/V sync issues, but its still not correct compared to v5.3.6
1 Like
This issue seems to be a very low priority, if any, for developers. Like I mentioned in the past, I lost my hopes long time ago to have this fixed. That’s why I still use v5.3.6 for interpolation, even though even that one is not perfect because it creates some random artifacts, especially on fast pano movements of the cameras or high speed action scenes.
1 Like
Deleted
Wait…Your compressing the output from Topaz in another program, and that program exports the video as variable framerate?
Correct. When setting “extract timecodes” to “always” in Staxrip, FPS metadata in output is adjusted accordingly like in the above example to 60.111fps (as well as retaining A/V sync since timecodes are correct in this case) which does not match the metadata that TVAI initially shows. If metadata matched timecodes then the FPS would be 60fps for both input and output regardless of whether I used timecodes or not, like with v5.3.6 and prior.
Using timecodes like this can often change the metadata from constant to variable given how very small decimal place differences (in timecodes vs expected 1/60 calculation) are flagged, so isn’t reliable at detecting variable FPS.. however.. splitting that same file into 2 parts and rechecking metadata shows a slightly different FPS for each file, so yes, the result is still variable FPS.
Staxrip atleast (without timecodes extraction) identifies that the FPS is not 60 because it does its own calculation of frames/length, but not sure why it only calculates using the first 200k frames though 
Calculating based on total frame count and length in seconds…
288497 / 4799.148 = 60.11421193928589 (average) FPS for a supposed “constant 60.000 FPS” TVAI output.
Using the default treatment of timecodes extraction (of extracting only when VFR metadata is detected, which many other editing software also follow), they would get discarded since TVAI output is incorrectly labelled constant 60fps, resulting in your newly edited or compressed file having the A/V going in and out of sync throughout the file. Of course, YMMV depending on your TVAI workflow, how you handle the TVAI output and if there’s any additional processing it undergoes etc.
I presume you’ve checked the integrity of the source files frame timing?
Yeah. Frame timing of the various source files I’ve tested is fine. The majority of them are 1080p remuxes in 23.976fps. 5.3.6 works fine for interpolating them, but has issues with some models having artifacting.
To fix an issue in the app, we have to be able to replicate the situation consistently so that the devs can understand what is causing the issue. This can take time to sort through, especially with complicated issues. Getting logs from users who are experiencing the issues can help speed up this process.
In the case of the situation you are seeing, it took a while for me to even replicate it and sort through the data to find that it only happened when using videos that were smaller than a certain resolution size and when the app was seeing them in non-square pixel format.
Our support team, the devs and everyone at the company cares about issues in the app and want users to have good experiences with our apps. There have been multiple changes since 1.0.3 and even before that release, that could be contributing to this issue as all of them are being reviewed to sort out what is going on here so that it can be fixed.
In my comment above, I stated I was able to replicate the situation with a file, and then by converting the file to a square pixel format, I was able to use the model as expected. Have you tried this approach and still run into the issue? If so, please send us the logs so we can continue to look into the situation.
The Devs can reproduce this inhouse without user logs. VFR or fake VFR exists since the change to mplayer…there were fixes, but was never 100% fixed, V5.3.6 is the latest VFR free version. To reproduce, take any video, i recommend 29.97 or 23.976 then it happens every time. Do frame interpolation from 23.976/29.97 to for example 60fps, keep “replace frames” on…result is VFR.
But it could also happen (not every time) when source is 30fps and output 60fps, you may can can see small difference in fps as described above, for example 60.1 fps out, maybe you can see it after re-encode or remux.
5 Likes
