In my Topaz Video Enhance AI adventures I’ve come across a need that I think Topaz’s AI know how much be able to help with restoration work.
While not so much an issue with modern film production, footage shot on traditional film media is easily subject to damage, be they quick spots of dirt or scratches or more severe damage (vertical lines, bits of the frame that have been scraped off from damage, etc).
Why not use AI processing to repair this damage using neural networks?
It would need to be temporally aware obviously but would be an amazing tool (assuming it could be done).
(I would attach png examples of the types of damage but I don’t have any that I don’t have the rights to reproduce on the forum right now.)
(NOTE: I originally posted this under General Discussions but realized this is probably a better spot.)
This has my vote, too. Although nearly all image content made today is digital, there are literally hundreds of millions of prints, slides, and negatives that could be digitized. More would be except for what a huge pain it is to remove dust specks and scratches on post-processing. I would pay $500 for an application that did a really good job of this on stills. Add another $500 for a product that would do the same with digitized film.
This would be a fantastic feature. Currently this has to be done mostly manual using very costly software. Archival film restoration is indeed a very nice market, hence the pricing. There are currently many people scanning their old 8mm films from the past using affordable and easy to use hardware like the Wolverine Moviemaker Pro. I would happily pay for this feature.
I actually came here today to ask for this. I digitize old 8mm and Super8 film which typically has dirt, scratches, color changes (fading/“reddening”) etc. Video Enhance doesn’t really help with those currently, although it does help with the grain somewhat.
I’ve definitely love a feature like this. I digitise negatives/slides using film scanners with ICE which does a pretty good job, but the ICE program is showing its age and is limited in its abilities.
Silverfast scanning software comes with its own version of ICE but their pricing structure on their software is so ridiculous that I haven’t bought it (you have to buy a copy for each type of scanner you have).
Having said that, they have released a program for removing scratches and dust marks from photographs where there was no negative to scan. It’s a plugin for Photoshop called SRDX. I’ve found it to be quite useful in some circumstances cutting down on the manual work required in removing dots, but other times it’s not so great. It’s also not been updated that much since release so I don’t think they are doing much to improve it.
I love the continual updates and improvements to Topaz software so if they could bring out an AI dust/scratch removal tool for still and, if possible, video, that would be absolutely amazing!
I would also love this - I’m currently sorting through some old family 8mm footage from post WWII Hungary. Apart from family value, there’s also some historical value to the footage and it would be so good to be able to clean it up properly.
There is professional software already available for film damage. I think it would be best to keep this separate from this software. On a cheap note – have you tried Neat Video? It does a good job in general and it is cheap.
Can you explain your reasoning for this in greater detail? We already have AI denoising, sharpening, masking, video enhance, etc… from this company. This post seems to simply be pitching an idea for an entirely new product from Topaz. Is that an issue somehow?
I’m sorry – I am just commenting on this product being used in a commercial situation. There are a lot of good tools out there for fixing film damage. Film damage is a very vague term also – what type of film damage?
In my opinion – that is a whole other world.
This product is designed for upscaling video at a higher quality and appears to be one of the top solutions out there.
Film damage software has existed for some time. Perhaps you are looking for a consumer version of software to fix film damage, which I doubt there is a big market for this – but hey if they add this extra toolbox then – God bless them!
My point is – why mimic or copy what already exists?
There are already a least three powerful products on the market, which are: Neat Video, Diamant-Film Restoration, and PFClean.
ABSoft is very known for years for its two products Neat Image and Neat Video. CUDA-capable NVIDIA GPUs and a large number of AMD video cards via OpenCL, and Metal.
Diamant-Film Restoration is really expensive. This professional software is one of the leading high-end film restoration solutions on the market. ONLY A demo is available on their website. To order their software, you need to contact them. For CUDA from NVIDIA and NVIDIA RTX graphic cards with a minimum of 8GB RAM - A MAC version is available.
PFClean is the tool of choice for key industry players such as Sony, Deluxe, Technicolor, The Criterion Collection, IMAX, Pinewood Studios, BBC, etc. Prices: 30 days: €399.00 / 90 Days: €839.00 / 1 Year: €2,299.00.
i’m actually thinking about buying neat video plugin for vegas pro. it’s expensive because of the VAT, but compared to these other softwares i wrote to the company but didn’t got a reply about their home / pro products. as i want to output 1920x1080, but process 1920x1440 file. so i don’t know if the home one work for that or not, as the pro one is double the price
i will “write to them” again… i need such software because i’m getting big block compression on videos in dark or with uniform background (concerts with lot of black background or uniform color) once processed in VEAI … and so need a way to remove them in big parts. some people helped me, even got a customised QTGMC script / command , but it must be adjusted for each videos, / source. my computer is too slow, so i’m trying to think about the Neat solution.
I think Video Enhance AI would probably be more effective at film damage repair if it was more temporal in nature (it’s current behavior seems to be treat each frame as it’s own image - not sure if 3.0 is going to change that or not).