Hello everyone,
I am experiencing a severe and persistent issue with Topaz Video AI (v5.x/v4.x) on a clean installation of Windows 11. The software freezes on “Estimating” for over a minute before starting any preview or export, and the logs are flooded with cyclical errors every second.
My Hardware Configuration:
-
CPU: Intel Core i5-14600K (14th Gen)
-
GPU: NVIDIA GeForce RTX 5070 Ti 16GB (Blackwell Architecture)
-
OS: Windows 11 Pro (Clean install, running a Local/Anonymous User Account, Windows Defender/UAC tailored for privacy).
-
Storage: Primary OS NVMe SSD, secondary Samsung 870 Evo SSDs for scratch/workspaces.
The Symptoms & Log Analysis:
When loading a video input generated via ComfyUI (H264/NVENC, 32 FPS) and applying a combined pipeline of Iris (Upscaling) and Apollo/Apollo Fast (Frame Interpolation to 60 FPS), the processing gets stuck at 0% for 60-90 seconds.
Looking closely at the .tzlog files, two critical behaviors emerge:
-
The
moov atom not foundandTypeErrorLoop:
FFmpeg goes into a cyclical loop every single second spitting out:
Format mov,mp4,m4a,3gp,3g2,mj2 detected only with low score of 1, misdetection possible! moov atom not found
This happens even if the input video file is completely healthy and contains no index errors. -
The Blackwell / NVENC / Lookahead Conflict:
Digging into the exact FFmpeg command invoked by the UI, I noticed that Topaz forces these flags for the Nvidia encoder:
-rc-lookahead 20 -spatial_aq 1 -g 30
Because the input video has an AI-generated variable/non-standard frame pacing (32 FPS via Rife) and Topaz forces a hardcoded GOP of 30, the lookahead function on the new Blackwell architecture (nvenc) goes into an asynchronous memory allocation stall. FFmpeg hangs, times out, stops writing the temporary preview buffer, and triggers the cascade of falsemoov atom not founderrors. -
The Local Account / Zip Extraction Bug:
The logs also explicitly show:
CRITICAL: Cannot open zip file at: C:\Topaz_Models\iris-v2-...tz
followed byCACHING MAY NOT WORK. When Windows 11 runs under a Local Offline Account, the security layer handles the temporary decryption of the encrypted.tz/.tz3model archives poorly upon system reboot. If Apollo or Iris try to open 2 instances simultaneously offline, the engine falls back to DirectML (DML with device: 0) instead of TensorRT, destroying initialization speeds.
The Only Working Workaround:
The only way to bypass this initial freeze is to completely avoid the MP4/AV1 Nvidia hardware encoders for previews/exports. If I switch the Default Export Container to MOV (ProRes 422 LT) or MKV (FFV1 8-bit), the lookahead function is skipped, the moov atom is structurally absent, and the RTX 5070 Ti immediately triggers the TensorRT cores, launching the preview in less than 2 seconds.
However, exporting huge ProRes/FFV1 files for short clips is unsustainable for high-volume workflows, forcing me to batch-recompress everything back to light H264 MP4s via Adobe Premiere Pro / Media Encoder.
Request to Developers:
Please look into how FFmpeg handles the -rc-lookahead and -g 30 GOP boundaries on the new NVIDIA RTX 5000 series (Blackwell) when processing non-standard framerate inputs, and fix the .tz model zip extraction stall under Windows 11 local user environments.
I can provide full log files upon request. Thank you.
