Yes exactly, I have to roll back to 7.1 to process this 10 hour video again. ehhhhhhh
I got an “AI Model Fail” evntually approx at the last hour from a 10-hour Proteus enhancement with 4k. So frasturated. I did notice the export info from the export queue, it shows several “downloading proteus-v4-fgnet…” once I hit “Export as” locally, even though I have Proteus model already downloaded from the Model Manager window. and it leads to “AI Model Fail” error with 10 hours wasted.
in my situation it eventually load yes, but it leads “AI Model Fail” at the last hour from a 10 hour process time, have to roll back to 7.1, and no loop downloading model anymore, another 10 hour process time then, eh
The new updates unfortunately have a few bugs, which are critical, as I can no longer rely on this tool to upscale. Here’s a scenario:
Upscale video x 2
Convert frame rate to 60 fps
Model = Iris, Low quality
Once converted it’s full of flickering artifacts. I hope you can fix this very soon, until then I cannot use Topaz Video.
Thanks.
I get V5.3.6 vibes, downgrade to 7.02 and stay there
I’ve been waiting for over 10 minutes to download Iris for preview and still nothing happening!
Aren’t updates supposed to improve the app?!
I uninstalled and reinstalled and models are not downloading!
Is Topaz using A.I. to code this software?
Then the A.I. has been trained with the sourcecode of previous versions of TVAI and that’s why the same bugs keep reappearing.
I’ve been mainly using Rhea, and I find this update has made my upscaling 25% slower. Also, every single video in the queue says “downloading model” 5 times over until it finally clicks in and starts the upscale. It happens at the start of every video.
Why does this version always have to re-download the models. Before it would download the models once and you were good to go. But this version, click 10 second preview, gotta re-download the models. Click on it again to render another preview, gotta re-download the models. Export the project… gotta re-download the models. Seriously, WTF?! I’m fed up of having to uninstall and roll back to an earlier version because of a bad update. I’m sick of this.
Bahaha! Behold, the true power of AI!
Great output, I have used starlight mini on some 1996 CG clips with 352x288 old CD, similar result 
That’s what I’m saying, man. Topaz has a monopoly on the AI video enhancement market though because no one is even close to as good as them, so they can continue to ignore their users and make changes no one wants while still profiting off of us because we have no other choice but to use their software if we want to enhance/upscale video.
Actually there ARE quite some very capable open source models out there - but installing / using them is a real PITA at the moment.
Also the hardware requirements are very high.
If someone wrote an app making the use of those models (SeedVR e.g.) a simple one-click “load video → export at 2x (or higher)” solution TVAI would have a hard time.
Using SeedVR isn’t as difficult as you might think. I started from scratch and am currently making my videos with it, even at 6-7 fps. There’s a saying: “Getting started is half the battle.” Detailed information on installation and usage has already been shared on another forum.
Adding text models to TVAI would be nice. It still distorts text a lot. There are actually free pre-made text models, and they’re very small, around 50-60 MB. Why don’t they add them to TVAI?
For example, you can check out this link:
Look how much text has been corrected. This site has thousands of free pre-made models for use in open-source applications like ComfyUI.
The ComfyUI Thread is here
The Seed Project looks great….
Good competition for TVAI.
Well, maybe not “good competition.” But at least some competition.
Because it’s entirely too niche. No way I’m going to spend the time trying to learn that set up and functioning. Tried once….wasted time.
If they put a solid, simple UI on the front of it and simplify….a simple install…then… “Click here get 2X AI enhanced video” it becomes something the community as a whole will consider.
Until then…..I will just use Starlight Mini with its slow as hell speed and spend my time on other things like…life.
It works….and works damn good on the stuff I am doing….that simple.
To add some humor to this, we should have a contest on who has the longest successful render in Starlight Mini. Winner gets a free kid’s meal from Mickey D’s.
My personal best is six days five hours. GeForce 5070 Ti, default settings. Average speed was .5 frames per second.
Anyone else feel free to chime in with your best/longest. I’ll bet there are some out there that will blow me away.
Yes, me too.
Yes, my 8K video enhancements with Proteus have dropped from 1.4 to 1.2 fps. For 15-hour jobs, that means it now takes me 2 hours longer… 
I’m experiencing the same thing. With Proteus, my speed has dropped from 1.4 to 1.2 fps. In my work environment, projects that used to take 10 hours now take 12 hours…
Since the last update there’s a problem with downloading, extracting and loading models.
2025-08-17 19-26-52.231 Thread: 34480 Info FF Process Output: 18
2025-08-17 19:26:52 43940 INFO: PARAM DEVICE: 0
:AIINFO: Downloading Model rxl-v1-fgnet-fp32-288x288-4x-rt1200-10800-rt.tz
2025-08-17 19-26-53.593 Thread: 34480 Info FF Process Output: 18
2025-08-17 19:26:53 43940 INFO: RESULT for https://veai-models.topazlabs.com/rxl-v1-fgnet-fp32-288x288-4x-rt1200-10800-rt.tz
2025-08-17 19-26-53.593 Thread: 34480 Info FF Process Output: 18 is 200
2025-08-17 19-26-54.921 Thread: 34480 Info FF Process Output: 18
2025-08-17 19:26:54 43940 INFO: Compressed Buffer for rxl-v1-fgnet-fp32-288x288-4x-rt1200-10800-rt.tz verified
2025-08-17 19:26:54 43940 INFO: putting fetched file into: D:\Topaz Labs LLC\Topaz Video AI\models\
:AIINFO: Model Downloaded
2025-08-17 19-26-55.472 Thread: 34480 Info FF Process Output: 18
2025-08-17 19:26:55 43940 INFO: Created temporary file at “rxl-v1-fgnet-fp32-288x288-4x-rt1200-10800-rt.tz” with 104071616 bytes
2025-08-17 19-26-55.481 Thread: 34480 Info FF Process Output: 18
:AIINFO: Loading Model
2025-08-17 19:26:55 43940 INFO: ModelBackend loading from file
2025-08-17 19:26:55 43940 INFO: CACHING MAY NOT WORK"D:\Topaz Labs LLC\Topaz Video AI\models\rxl-v1-fgnet-fp32-288x288-4x-rt1200-10800-rt.tz"
2025-08-17 19:26:55 43940 INFO: Reading model file duration: 0.17 ms
2025-08-17 19-26-55.491 Thread: 34480 Info FF Process Output: 18
2025-08-17 19:26:55 43940 INFO: TRT loading model from rxl-v1-fgnet-fp32-288x288-4x.trt 121526444
2025-08-17 19:26:55 43940 CRITICAL: TRT Issue ERRORIRuntime::deserializeCudaEngine: Error Code 1: Serialization (Serialization assertion stdVersionRead == kSERIALIZATION_VERSION failed.Version tag does not match. Note: Current Version: 94, Serialized Engine Version: 239)
2025-08-17 19:26:55 43940 CRITICAL: Can’t deserialize CUDA engine
2025-08-17 19:26:55 43940 CRITICAL: Loading default error reading file: D:\Topaz Labs LLC\Topaz Video AI\models/rxl-v1-fgnet-fp32-288x288-4x-rt1200-10800-rt.tz
:AIERROR: Model Failed
2025-08-17 19:26:55 43940 CRITICAL: Unable to create or load model of type tensorrt from path: rxl-v1-fgnet-fp32-288x288-4x-rt1200-10800-rt.tz
2025-08-17 19:26:55 43940 CRITICAL: BackendRunner cannot be initialized for model: rxl-v1-fgnet-fp32-288x288-4x-rt1200-10800-rt.tz
2025-08-17 19:26:55 43940 WARNING: Failed to load model rxl-v1-fgnet-fp32-288x288-4x-rt1200-10800-rt.tzfgnet-fp32-[H]x[W]-4x-rt[C]-[R]-rt.tz[“out”]
2025-08-17 19:26:55 43940 INFO: Blocks total: 1 N 18 Blocks count 18
2025-08-17 19:26:55 43940 INFO: Checking block sizes 1956x1056 penalty 0.05 overlap 48
2025-08-17 19:26:55 43940 INFO: Make backend 12 input size 1956x1056 block 384x576 onnx16 device 0 memory 1
2025-08-17 19:26:55 43940 INFO: Computing device instances
2025-08-17 19:26:55 43940 INFO: Apple 0 DEVICES 0
2025-08-17 19:26:55 43940 INFO: DEVICE INFO 1
2025-08-17 19:26:55 43940 INFO: Creating model name rhea instance 000002326007BEB0
2025-08-17 19:26:55 43940 INFO: NETS [“fgnet-fp16-[H]x[W]-4x-ox.tz”] OUTPUTS [“out”]
2025-08-17 19:26:55 43940 INFO: COMPUTED FILE NAME fgnet-fp16-[H]x[W]-4x-ox.tz rxl-v1-fgnet-fp16-576x384-4x-ox.tz
2025-08-17 19:26:55 43940 INFO: Locating local model: “D:\Topaz Labs LLC\Topaz Video AI\models\/rxl-v1-fgnet-fp16-576x384-4x-ox.tz” status:
2025-08-17 19-26-55.492 Thread: 34480 Info FF Process Output: 18 1
2025-08-17 19:26:55 43940 INFO: CACHING MAY NOT WORK"D:\Topaz Labs LLC\Topaz Video AI\models\rxl-v1-fgnet-fp16-576x384-4x-ox.tz"
2025-08-17 19-26-56.844 Thread: 34480 Info FF Process Output: 18
2025-08-17 19:26:56 43940 INFO: Loading filename rxl-v1-fgnet-fp16-576x384-4x-ox.tz fgnet-fp16-[H]x[W]-4x-ox.tz [“out”]
2025-08-17 19-26-56.844 Thread: 34480 Info FF Process Output: 18
2025-08-17 19:26:56 43940 INFO: Total Devices: 1
2025-08-17 19:26:56 43940 INFO: DEVICE DETAILS: 0 INSTANCES 2
2025-08-17 19:26:56 43940 INFO: PARAM DEVICE: 0
:AIINFO: Loading Model
The errors repeat randomly up to 20 times before actually loading the model.
Please solve this, I usually batch 20 clips at the same time, and it’s now taking forever.
Currently in beta: Starlight Sharp
Probably, due to complexity slower than Starlight Mini id guess.
