Topaz Video 1.0.4 - Patch

Zeromus · November 22, 2025, 1:08pm

Open cmd window
cd \%TVAI_MODEL_DIR%
execute RUNNER.EXE

Here’s what happens on my computer:

C:\ProgramData\Topaz Labs LLC\Topaz Video AI\models>runner
Runner: started
module search path: [‘C:\ProgramData\Topaz Labs LLC\Topaz Video AI\models\DLLs’, ‘C:\ProgramData\Topaz Labs LLC\Topaz Video AI\models\Lib’, ‘C:\ProgramData\Topaz Labs LLC\Topaz Video AI\models\Lib\site-packages’, ‘C:\ProgramData\Topaz Labs LLC\Topaz Video AI\models’, ‘C:\ProgramData\Topaz Labs LLC\Topaz Video AI\models\Lib\site-packages\starlight_utils-0.2.0-py3.12.egg’]
Runner: Authenticating
platform: Windows
Machine ID: 3d8d0ddb-1da4-4316-bf91-f60994892901
auth file path: C:\ProgramData\Topaz Labs LLC\Topaz Video AI\models\auth.tpz
platform: Windows
Contents of auth.json:
b’{“auth”:“eyJhbGciOiJSUzI1NiIsInR5cCI6IkpXVCJ9.eyJtYWNoaW5lX2lkIjoiM2Q4ZDBkZGItMWRhNC00MzE2LWJmOTEtZjYwOTk0ODkyOTAxIiwic3RhdHVzIjoiYWN0aXZlIiwibGljZW5zZV9leHBpcmF0aW9uIjoiMjAyNi0wNS0xNSIsImFwcF92ZXJzaW9uIjoiNy4xLjAiLCJ1c2VyX2lkIjozNjcwNDE3LCJ0aWVyIjoicGVyc29uYWwiLCJmaXJzdF9uYW1lIjoiSmFtZXMiLCJsYXN0X25hbWUiOiJTaHJpZGVyIiwiYW5udWFsX3N1YiI6InRydWUiLCJpc3MiOiJwbHV0dXMiLCJzdWIiOiJ0b3BhemxhYnNAamVkaS5kb2ciLCJhdWQiOiIxNDY0OTUxIiwiZXhwIjoxNzYzODI3MDU3LCJpYXQiOjE3NjEyMzUwNTd9.sCwRYBfsqzAh4x3wQljr2mOfj4H_I3_G7aoC9CN6yIb_3HghBFECixJZs00cxCscWMdgJTLMEvvbNBpk3jbPuC4j9cCiOkGPL6E0_iBBh3z7Lp_a-qNYQ_2Seun7HL6TvtoL2NmWRL15lk3L81iUqeX4pU7bykBtQckagsvgnTX4SGUn9pY5tL2L6jyIjciwo2Av-MiLcbGDa9w7enxExe3VVdYu_T_KRWSUKfh94bJ0AWy9mWfpHE6sFuWS0_NicHwWoD-IHwuxQ4gEwwfzQ6hUGuKpRx6wY5ELXeKOP8p1NXoNpYb2SZEJ5_MPMZBXMCOvKBW7hyoz7HSVDCXwcxzoHO22vsnQpfidcaPkiwYgdpR2smh3GRNcd8rzBECkH6qbJ-GXO-euESWUiBurXuu8kZ3vtsmcdYSIVYcXaKhB8J5cUfahax-T3s7u141AiRRSMLxmrwmghNSccazLV5uJxVCJP9U9O71apMp6ETRhBAj9mwQZOD8xCgAtE3spEIyTWNA55uZ-fGUhOe3oBNnjMRecimscs46cuGGbc7QLth_hKDlXFiIi9AYqz6xjGdtdJPQhE3pQTPdDnknYughX158XqObo9-RtoupXFD5GqCeGSlyu1xIB64Fzbr4AEvz5HKPStJ1-CFCD0BxVextUv0HkZqqRRLtQ_sRm1CQ”,“expiration_date”:“2026-05-15”,“issued”:“2025-10-23”,“pro”:false,“refresh”:“eyJhbGciOiJSUzI1NiIsInR5cCI6IkpXVCJ9.eyJtYWNoaW5lX2lkIjoiM2Q4ZDBkZGItMWRhNC00MzE2LWJmOTEtZjYwOTk0ODkyOTAxIiwic3RhdHVzIjoiYWN0aXZlIiwibGljZW5zZV9leHBpcmF0aW9uIjoiMjAyNi0wNS0xNSIsImFwcF92ZXJzaW9uIjoiNy4xLjAiLCJ1c2VyX2lkIjozNjcwNDE3LCJ0aWVyIjoicGVyc29uYWwiLCJmaXJzdF9uYW1lIjoiSmFtZXMiLCJsYXN0X25hbWUiOiJTaHJpZGVyIiwiYW5udWFsX3N1YiI6InRydWUiLCJpc3MiOiJwbHV0dXMiLCJzdWIiOiJ0b3BhemxhYnNAamVkaS5kb2ciLCJhdWQiOiIxNDY0OTUxIiwiZXhwIjoxNzcyMDI5MzM2LCJpYXQiOjE3NjEyMzUwNTd9.ejTLs38W9J2iG6159D13aiTqvxYD_5rdaALMqPwotjAfg6PaipHHQuqM0g_3UGaU4iTBmM0kr6cCo_dfyaHoo-oViUa9iOuG3qxSQRuzdv6U0oAgL3XuDm2Tr6DwhIGYm5DF_kTmRc340Rnaw_E8i3iwgip7hY4aMPWpnkcDhpb_NZ6wOo22jCwrhL5YFuBgT2ccmDSXNtwAjBWzCKu7YyFmJkBOWmVRDfOVWf5viPHyKKezu3Ica_w_qKgXO5tlLRGThpJ2t9WID-VGgZ0oo8-KLOC2_NbF1-ZQhYJwiw1JB0KYZ8eUuRIvwkqTZw7xJ10nV0xjhSYimbziEyLTBUUQi_jjVO0rdgFXCCjVhnNiAHcX6eDB7_rJgnoGQ5pIoc5D6fOJ3FF1FiRd837IWMTCdhAEEUuo3xdXKadFJKvs1Z6CVNMpoyB-J4wrGjjPfaUva0ajumHJXeHhBH-HgB0TxBqEFibEJiP6DLqOcECvyNyvHC_WdAstQV35GHFAG-QXnTsrA5GVMdo455NnoLlO35jucUkJVkRwc3Glj7Mpgy6Wz4YJbnRiXkEWzIQs-uBzc_ORKiHI9ZQ7K4LYtXEV0ShvRYwAyaEnTZlwmZO5TTeSnWnh0U8MhcFRDiCjRUW4L6kJRZEujjupmIBobJ8djRrrnFd4qbebBZuRt3s”,“refresh_expires”:“2026-02-25”,“relogin”:“2025-11-22”,“status”:“OWNED”,“username”:“*********@***.***”}’
Authentication token expired
Will attempt to refresh
Reauthentication successful
Refresh succeeded. Reusing existing license.
License Date: 2026-05-15
User Email: *********@***.***
Is Owned: True
Is LoggedIn: True
Auth Check Watermark: False False False 2024-04-01 2026-05-15
Runner: main
usage: runner [-h] [–start-frame-idx START_FRAME_IDX] [–end-frame-idx END_FRAME_IDX] [–upscale-factor UPSCALE_FACTOR]
[–ffmpeg-encoding FFMPEG_ENCODING] [–device DEVICE] [–max-gpu-mem MAX_GPU_MEM]
[–ffmpeg-preproc-filters FFMPEG_PREPROC_FILTERS]
input_path output_path
runner: error: the following arguments are required: input_path, output_path
Runner: exit code: 2
Runner: Finalizing
Runner: Done

C:\ProgramData\Topaz Labs LLC\Topaz Video AI\models>

What output do you get when you do the same on your computer?

ole.solev · November 23, 2025, 7:03am

Everything is fine now. Thank you.
I didn’t copy all the files from the DLL folder earlier.

michele.meschke · November 23, 2025, 12:14pm

YES…i can confirm that Topaz Video Frame Interpolation with Apollo V8 with SceneDetection is totally unusable !!!
The exported Video is about the double in time length…
At the scene changes the video is just freezed in for some seconds….video and audio is going totally out of sync.

maybe without that new feature “Scene Detection” it´s working ?

Result: Scene Detection is extremely slow and the result is just shit….Without “Scene Detection” it seems to work….

Zeromus · November 23, 2025, 1:52pm

You’re welcome. I’m glad you got it working.

robert.roland · November 23, 2025, 7:48pm

I am experiencing the strangest problem with Starlight Sharp and my GPUs.

TVAI runner.exe is being assigned to my slow GPU., but ONLY when using the SLS model.

I have ensured that the TVAI GPU is set to my 5090 as it always has been, and in Windows 11 graphics, runner.exe is assigned to the 5090.

I have tried restarting etc., but the behavior stays the same.

Even more strange is that when you look at Windows task manager it shows runner is using the correct GPU, but if you look at HWiNFO (and GPU fans and est. time to completion), it is obvious the slow GPU is being used. The 5090 is idle primarily, with occasional jumps to 100%, but the slow GPU is 100% pegged.

Again, this only happens with SLS. SLM works fine as it always has and uses the 5090.

Rhea, Proteus, etc., all use the 5090 as they always have.

This makes SLS unusable for me, and I wonder how this is even possible?

I don’t recall this behavior when SLS first came out. But I usually don’t use it, so I can’t be 100% sure about that.

Anyone know anything about this?

TheRustyRebel · November 23, 2025, 7:57pm

Since Starlight Mini/Sharp rely on GPU memory size, would a 3090 get better results than a 4080 card? Since one has 24GB & the other only 16GB.

robert.roland · November 23, 2025, 10:26pm

I think both cards would fallback to the 13GBish (or even 8GB) usage, meaning the GPU processor might matter more. The quality difference between a 12-13GB run and a 23GB run is negligible to me.

On my setup, SLM never allocates more than 23GB, even though I have 32GB, and have set the allowed allocation to 96%, and have nothing else running on the card.

Before the SLM memory issues were fixed a while back, I had to set the allowed allocation closer to 70% and my runs would use ~13GB. The quality was still stunning.

Back then, if I went over an 80% allowed allocation, and SLM consumed the 23GB, I risked falling back to DRAM, which made SLM go from 1.0fps to 0.2fps. SLM has gotten better, and with some tricks I can usually get 1.6fps-1.8fps using the 23GB and have no DRAM fallback.

You might get SLM to use the 23GB on the 24GB IF nothing else was running on the card (like your OS display) and you allowed SLM 98%. But you would probably fallback to DRAM at some point and kill the performance.

Check how much VRAM is being used just to run your display and a few apps. If whatever you have leftover is below 14GB, then you might want to go with the 24GB.

I think the next fallback VRAM allocation for SLM (when the 13GB is not possible) is around 8GB.

I would think both cards would run SLM around 0.2fps-0.5fps, depending on resolution.

robert.roland · November 23, 2025, 10:36pm

I have a setup like this. Did you make sure your BIOS will allocate your lanes (bifurcate might be the right term) to 8x and 8x instead of the 16x to the 1 primary GPU?

pcie4 8x might be a 10% performance drop vs pcie5 x16.

I keep my second card outside of the PC case to keep the temperature in check. I can’t imagine 2 high-powered cards running so close together they are practically lying on one another…

Dedicating a card only to TVAI will definitely speed up performance and free up more VRAM. In my case, it freed up around 3GB on average. My TVAI performance improved 10%-20% even with a 3% loss due to lane splitting.

robert.roland · November 23, 2025, 10:39pm

I just noticed you wanted to use 2 of the same cards for TVAI processing at the same time… only a pricey TVAI pro version will do that. I think you would need an NVLink cable also.

Save that money and just get 1 monster card.

What I did is just dedicate a card to my display and a card to TVAI.

It did help.

Moebius · November 24, 2025, 1:45am

I hope the starlight preview gets fixed at some point. My video is 30% done and I still don’t see a preview. I need to know if it’s going to look good, especially if it takes so long.

MVFX · November 24, 2025, 11:30am

I’m so glad I cancelled all my subscriptions. Honestly, everyone should. We’re being treated like test dummies while they keep taking our money and giving us nothing in return. No improvements, no meaningful updates - just promises and empty talk. Even my TV AI 7.4.4 wasn’t able to handle a fookin’ frame interpolation while Apple Compressor done it in minutes. I’m done with this CRAP!

jo.vo · November 24, 2025, 12:26pm

Or get two cheaper cards in midclass PCs and use your old two-seat lifetime license to do simultaneous rendering of two movies (or one movie split in the half).

Or even three PCs and use the old license on two of them and your Studio version on the third.

LowFlight · November 24, 2025, 2:29pm

You mean test dummies for 5 years, don’t you? This software has never amounted to what the marketing hype promised it would be. Even it we’re to make some miraculous turnaround, I’ve not only been priced out because of the strict hardware requirements, but also the high rental cost. I refuse to pay rent for software, of any kind. Think about it, all it will take is another major internet outage, like the recent Cloudflare disaster, to be locked out of your software.

robert.roland · November 24, 2025, 10:22pm

agreed

robert.roland · November 24, 2025, 10:56pm

I really think some on this page have forgotten what options people had to make video look better just 5 years ago.

This stuff is not easy.

Ok, maybe the price feels too high, but that is a function of very personal views of value and what the market will bear. More, low-priced competition or an even more amazing Topaz will sort that out quickly. Clearly, some of the pricing has been head-scratching, but some has seemed totally reasonable.

Hopefully, the revenue is being put to good use. It seems to me that there is an abundance of work to be done - and, therefore, plenty to spend that revenue on.

Don’t get me wrong, I hate subscription pricing too. Not long ago, I went ballistic on Adobe to enforce my perpetual licenses until they finally relented. But, to me, this is a different situation.

Topaz is much more of a startup and needs some stability to make the heavy lift. Adobe, at the time, was already a giant monopoly that did it greedily because they could back their users into a corner. Right now, at least, people do have a few alternatives.

I’m reluctantly ok with it, as long as the early adopters were/are grandfathered. At least those who recently bought the perpetual versions, which are quite good, have been promised they own them forever. And when the subscription came around, you were offered a big discount that gave you their entire suite - not just what you paid for.

For the haters, maybe you haven’t worked in a startup to understand the myriad of challenges that include what’s going on. Or maybe you have, and serendipitously, every pothole was avoided.

As long as Starlight is working, I would be happy with that for quite a while.

LowFlight · November 24, 2025, 11:42pm

Hello Robert,

I paid all the way up from version 1 to version 3, and it never performed like the Topaz marketing said it would. I’ve upscaled some old 1960’s sitcoms, but the results are nothing like content originally captured in glorious 4K resolution, as was the promotional marketing at the time. I feel I got suckered. Now, with that said, it doesn’t mean I’m not happy with other Topaz products. Gigapixel helped me upscale some very old photographs from the early 1900’s to a resolution that I could have professionally printed and frame. For me, Video AI was a scam, to keep folks chasing the carrot, but I refuse to chase anymore carrots. And the nail in the coffin came with subscriptions. Adobe lost a vast majority of their home users and hobbyists when the went to subscription model, but at least they allow up to 99 days in Creative Suite before you are required to call home. When comes right down to it, software subscriptions are a way for companies to overcharge for software you use on a regular basis. If your not earning money with that software, then it’s a waste of money.

Mayday · November 25, 2025, 7:18am

The crazy thing is that Topaz often creates its own problems. They invent a new UI that nobody likes, which they could have saved themselves the trouble of doing. They create a “Project” function that annoys people, we don’t ask for that.

You break features that worked before, and this is crazy how often this happens. No one tests properly before something is released. You make poor use of the community’s power; many here are willing to help, points out bugs and give improvment tips, there are also some cracks here pointed out ffmpeg faults, but often there is no reaction or "we are working on it” yes sure…

What is needed now is just a single person from Topaz Labs who makes a list oft all bugs, traces and documents everything in a test environment, so that it can then go into development where the bugs can be fixed and improvment can be made, after which it will be extensively tested before release.

All existing functions should also be reviewed. For example why not every model has add noise/grain function? This could be added to all models and by the way they could improve add grain with random grain.

For models that are freely scalable, such as Starlight, you could also give users freely selectable upscaling factors. Topaz is not making full use of the possibilities it has.

SkodreragGaming · November 25, 2025, 8:16am

I have a say too: I have a say too: I have been warning you for years about all of what was going to happen and no one listened to me. I was just told it’s not true or it’s false. Well Well the result is there today which is completely catastrophic. I told you all that since the change of team at Topaz at the end of version 2 it would be a catastrophic in the future. Well, we are right there. This started since the team switch after the end of version 2 which was already very stable and which may not need new versions. The change of team was made from version 3. For my part, I find that the old version 2 team was much more competent than the one who took to do version 3.Many people on this forum made fun of me because I was telling the truth and now you have the truth in the face today. I think that within a month I will cancel my annual subscription for this software which is no longer the one I knew when it first started. I think to remedy all of this is a deepening change from the team behind this software.But, as to what extent will this change of team solve the problem, it may not be safe. But I simply say that the team from version 2 was competent and it was really listening to people on the forum they managed to solve software problems immediately, which made the latest version 2 much more stable than the new versions of today. As I will say, return the old version 2 team to us who were competent and especially attentive to people, but I know in advance that this will not happen.I am simply saying that Topaz is heading straight towards the wall and at the bottom of the ladder. There I also gave my opinion. Maybe some of you won’t agree with what I’m saying and others will agree.

dianac · November 25, 2025, 6:11pm

Preview works fine if your output resolution is 1080 or less. I am processing 480p VFR originals, so I process them first through ffmpeg to convert to CFR (@29.97 fps) and upscale to 540p. Starlight then offers 1080 as the minimum output. That gets the preview working, even with sound.

david.123 · November 27, 2025, 3:38am

I don’t think there was much of a team when v2 came out, maybe only a couple devs.

And v2 isn’t that good. There have been models that have come out in the 3+ series that are better than anything in v2 (iris, proteus v2, apollo). Features such as retain detail, and being able to add grain and noise.

The underlying architecture is far better now. It can become better still. I’ve had many direct conversations with the lead AI engine dev, and he knows what he’s doing.

I think the bigger issue is the almost frantic nature of the constant UI tinkering that adds nothing useful most of the time, and in some cases makes things worse.