DID I MISS SOMETHING? STILL NO ALPHA CHANNEL IN 3.2.X (!?)
NO BUENO!
Yes. You missed the part where this has never been an intended use for TVAI and they would have to make all new models to be able to process them correctly.
Okay. Thatās just my opinion based on all the marketing and comments I have read about TVAI.
If anyone from Topaz were to officially say they are going to add Alpha channel support. You can expect it to take about a year and a half form that time.
Sorry to be so negative. Your expectation seems unrealistic to me.
Alpha channels are used heavily by all post-production studios since they are essential to the compositing pipeline. Itās an industry standard and saying otherwise is honestly laughable.
Intended use? Channeling Topaz re we? I am not aware of one serious industrial strength digital processing app that does NOT preserve apha channel. Sorry but it would appear that youāve tripped on your premise and sprawled on your conclusion.
Your comment that Topaz should ignore this essential element because itās too difficult and/or would take over a year(?), I frankly find as irrelevant to the discussion. So, I suggest that IF Topaz wants to remain competitive in the marketplaceā¦that they best get startedā¦sooner than later.
As eloquently stated below by jason.fletcher, alpha channel is not some mere inconsequential element of post-prod but for serious compositing, it is an essential yes, indespensible element in the post-prod chain.
If Iām wrong. Iām wrong.
Yes. Iām am just a person that will only ever use TVAI to upscale their movie collection.
There is no such thing as a blu-ray or DVD that has an alpha channelāthat I know of.
What I do know more than average about, is what is required to train machine learning models. If I am right about how they have made what they have right now, than I would also be closer to being right about the time it would take to add alpha channel support.
I think the most likely outcome would be that a new AI model option is added that preserves and upscales the alpha channel. My rough guess is that it would be a variant of the Proteus model.
Thinking about the work involvedā¦ They would need to train a model to look at the red, green, blue, and alpha channels, which likely isnāt a crazy request from the programming side of things. But they would also need a dataset which has alpha channels included, which might be the hardest aspect to curate. Iām unsure if they train using images and then have some temporal smoothing that they somehow incorporate so as to have the model ingest video. But if they train using video then thatās more difficult to curate the dataset. They are clearly training models all of the time and I expect that they have several GPU racks on-prem, which is much cheaper in the long run than working in the cloud.
Overall the bottom line is, if they donāt add alpha channel support then someone else will and those of us in the VFX industry will check out that tool instead.
But at this rate it seems likely that Adobe will release their own uprez tool within After Effectsā¦ Just a guess though.
Iām really impressed with Topaz video. Itās been a part of my tool kit for about a year now mostly for if I want to clean up video. However I also do a lot of work as a CG artist. This weekend Ive been up against a deadline so instead of rendering my CG at 4k I exported HD which takes a quater of the time and then up scaled the video to 4k in Topaz video.
The results are flawless but I had to upres the final comp as opposed to my CG renders because Topaz does not support alpha channels. This makes the workflow pretty awkward and limits what you can do.
I can imagine that upressing alpha channels so that they still work perfectly without weird edges is quite the task. If you were to pull it off though then this is a tool that could completely up end CG workflow for all studios and individuals. Topaz video is currently super useful in a videographers tool kit. If it carried an alpha channel I believe it would quickly become standard use in CGI pipelines too.
What would be really incredible would be the ability to input a multi track .exr and upres/denoise that for re output to multi track exr. That would be drop dead show stoppingly amazing and I honestly think every CG artist in the world would want to buy that.
Video Game textures as well for alpha, I had to seperate into RGB for TPAI, then used a simple gaussian interpolator for the alpha, then had to merge the planes back together. For cutout gradient alphas it lends to mismatches at edges with sharpened colored edges going to black bleeding thru.
Itās not designed with our workflows in mind but with a bit of multichannel support it could be! Things like spec and diffuse wont be a problem but I think getting alphas to match would be much harder. Perhaps a bit easier if not attempting to up res baked in motion blur.
What weāve been doing is upresing with topaz and then using the source video as a matte for the alpha but having alpha straight out of Topaz would be amazing
Yes, please, alpha Support, would be a banger!
Alpha would be awesome!
BUMP !!!
Alpha is main request when you upscale/cleanup 3d rendering and today also if we have powerful cpu and gpu, 3d rendering slowdown be cause we need to render more, more resolution, more gigapoly and Implementation of Alpha in Topaz should be a good thing.
If you not do immediately, allow to do a dual render for RGB and Alpha sequence where algoritm see how to upscale from RGB to optimize upscaling of alpha
Would TVAI really be faster than a render farm? You cannot use TVAI in a render farm setup, but you can use every 3D rendering software Iāve ever seen in one.
I may be off on the numbers, but Iām pretty sure you can buy an old mining rig for less than the price of a used RTX 4090, and it will still render the 3D video faster than TVAI can on a used 4090.
This idea is moot if each video card in the farm needs lots of VRAM. Itās been awhile since Iāve had my render farm. Like 512MBs of VRAM was normal at the time.
Alpha support added in version 5.2
Video AI can now copy alpha layers from inputs and merge them back with AI model output videos. Alpha layer copying is supported in the following codecs:
- QuickTime Animation
- TIFF
- PNG