Video Denoiser Best Practices
The video denoiser should be the first effect in the video effects chain, but due to heavy processing overhead playback speed is greatly diminished. To remedy this obstacle, I typically place Red Giant’s Denoiser III in an adjustment layer above the video clips, so I can enable or disable at will.
I have encountered inconsistent results when rendering, at times dissolves don’t properly work and sometimes a clip does not appear when the adjustment layer is enabled, or other effects don’t properly work. Many times all appears to work, as expected, then a clip doesn’t contain an applied effect.
If Denoiser is applied at the clip level, then enabling/disabling is inefficient due to the time required.
I have considered applying denoiser when footage is first imported and transcoding before beginning a project. The downside is not knowing what material will be selected and losing quality in the transcoding process, unless transcoding to an Uncompressed format, such as MOV/Animation.
It’s not that I’m new to Premiere Pro or After Effects, having used them almost daily for a number of years, I’m just finally getting around to asking for opinions on how to solve this on going issue.
How is this process handled in the professional arena with feature films or heavy effects centric projects?
Forgot to mention, I'm using CC2017
I confess I'm puzzled by the heavy use of denoiser.
Properly-exposed video shouldn't be noisy.
KGAN (CBS) & KFXA (Fox) Cedar Rapids, IA
Dave, I do a lot of weddings and have no control over lighting so dark venues = noisy footage, then DeNoiser can be a real life-saver in post! I don't disagree with you about properly exposing footage of course, but not always an option unfortunately.
Joseph, my workflow is to convert noisy footage before doing any editing. Use a quality intermediate codec and there will be virtually no "visual" loss in quality. For instance, use GoPro Cineform YUV 10-bit (under QuickTime presets) or DNxHD. Either of those will make a large, high-quality file.
I use them all the time for compositing and such, doing multiple generations without visual loss. Once you have that cleaned-up de-noised file, you can edit away without the system bogging down! And of course no weird render issues from too many combined effects.
Another benefit of having transcoded the footage is that these codecs are less-compressed than most camera-native codecs like AVHCD thus requiring less CPU power to decode the footage, so that also improves editing performance. Assumed you have adequate hard drive space for the larger transcoded files.
Safe Harbor Computers
Thanks Jeff for your speedy reply.
I'll test which codec best serves my workflow needs and plan for extra storage and time for pre-editing transcoded material.
I've favored MOV/Animation due to the uncompressed advantage. Since I don't anticipate viewing in a theater, one of these suggested codecs may be the right approach to eliminating the denoiser rendering overhead during the editing phase of POST.
For a current project, the distributor mandates the delivered material be in the ProRes 422 HQ format, which unfortunately isn't licensed to Adobe by Apple for the Windows environment. My work around is to use Scratch to produce the ProRes from an uncompressed MOV/Animation render. As you can imagine a 20 minute master is 160+ GB, whereas the ProRes is 26+ GB.
Thank you for your confirmation on pre-editing transcoding as a legit workflow method.
In regards to "I've favored MOV/Animation due to the uncompressed advantage."
There is no advantage - since cameras all compress the heck out of the footage to start with. A majority of the image information that originally hit the camera sensor is discarded before the video ever gets to the memory card. Converting such compressed footage to uncompressed cannot retrieve any of that lost data, so it really is overkill to create those gigantic files.
So why not just convert to H.264 then to save space? Because of the high compression rate, even more picture information would be lost with each conversion (generation). Therefore we find the compromise between camera-compressed footage and uncompressed, which is a good intermediate codec like Cineform or DNxHD.
The file sizes are manageable while the new file is visually indistinguishable from the original footage. Uncompressed would give the same results, but the file sizes are outrageous, thus the intermediate codec.
Safe Harbor Computers
My logic for "I've favored MOV/Animation due to the uncompressed advantage." is that the uncompressed copy would maintain exactly the quality of the original footage without any further compression. Certainly, one can't restore discarded image information, lost info is lost info.
Storage space is not an issue for me, so I don't favor h264 as a candidate for my transcode medium, due to higher compression than the codecs you recommend.
Interesting sidebar, I have had occasional audio sync issues when rendering Premiere Pro projects in Media encoder (all material verified to be 23.976) and yet the same exact project renders In-Sync as a Premiere Pro Export. Each render was MOV/Animation, with Red Giant's Denoiser III, max color depth, and max render quality. These renders were the Master of the final project to be imported into "Scratch" to create the ProRes version for the distributor. I then rendered H.264 from the ProRes were my check prints, so to speak, to verify quality on my end and to preview on Vimeo.
It seems that if I modify my workflow to incorporate your suggestions on denoising an intermediate before editing, I will save overall storage space, gain faster final render times, and eliminate AMEs possible sync issues. Though I have to test which of the two suggested codecs to use as the import codec for "Scratch" to create the ProRes version for the distributor.
You recommend CineForm and DNxHD, in your experience what advantage is one over the other?
Any further constructive observations would be greatly appreciated.
Thank you for your continued help in solving my initial workflow question with denoiser.
Just to be clear, I never use H.264 as an intermediate and never recommend doing so under any circumstance, is only for final delivery.
About uncompressed versus Cineform, DNxHD, ProRes...these are "visually lossless" codecs for the most part, meaning mathematically some compression may show up if you cared to analyze the footage somehow, but your eyes cannot tell the difference and that is really what we are talking about - will I SEE any degradation using this route? The answer is no.
Back to DeNoiser, you are correct that if you do that step first, everything becomes easier after that. The less-compressed intermediate codec edits easier, you are not fighting the denoiser in Premiere that bogs down editing, and final render is MUCH faster since DeNoiser is not being rendered also at that point. Just a win-win-win situation.
Another plus is that there can be unforeseen side effects when combining DeNoiser software with other effects filters - which should be applied first? Does it make a difference? Will it cause encoding errors when combined with other effects? Just removes a lot of what-ifs to denoise first, then edit.
Sync issues - I haven't used MOV/animation before so I don't know if that's an issue or not. I also never use Media Encoder, just encode right from Premiere, but that's just my workflow, not right or wrong.
Cineform versus DNxHD? I haven't tried to compare them directly for size and quality. Both should provide excellent results. Cineform is a .mov file so that may play into compatibility with other apps, while DNxHD creates an .mxf file which may not be as friendly with other apps. Create a short sample of each using same source and make your own determination.
I'd never heard of Scratch and a web search turned up something from MIT. Is that what you are using? If they are offering ProRes encoding on the PC, it's probably using FFMPEG underneath since Apple does not support ProRes encoding on ANY PC software, has to be a hack/reverse-engineered solution basically if someone does it.
Safe Harbor Computers
Your last point about "Scratch", it is one of the few products Apple has licensed to export ProRes. I could not afford to purchase it at $650.00 per year, so I opted for a monthly subscription of $79.00. It is designed for motion picture workflows.
I plan an extensive comparative test and will employ the results on my next project, which will be shot with a Red Scarlet camera, with new headaches I'm sure.
You have been very helpful and I greatly appreciate your help.
Interesting about Scratch being licensed for ProRes, have never seen that mentioned before and I've seen a TON of threads wanting to know how to encode ProRes on the PC and no one ever brought it up before. Good to know.
Assimilate really, seriously needs to work on their marketing though. I thought "Scratch" sounded familiar, perhaps related to Smoke and Fire, but those programs are out of my league. So I did a Google search, and what comes up is a Scratch editing software put out by MIT (the higher learning place...). Assimilate Scratch must be a closely-guarded secret for some reason!
Safe Harbor Computers
I stumbled onto Scratch by my quest to find a reliable Windows/ProRes solution. Previously, I used an in-expensive conversion program (Borsoft) which uses a reverse engineered process. For this current project, the distributor requires ProRes, as the delivery format, and I just didn't want to chance any issues, especially when in the media metadata (I use MediaInfo to check) there was some off the wall name next to ProRes.
I found out that Scratch was license from Apple, so my confidence level increased. It has much more under the hood than transcoding, but for now this one feature is my main focus. The non-standard interface and non-intuitive workflow requires some careful attention to their tutorials and help guides. But after one or two conversions these issues are in the rear view mirror. The choices of media codecs begins with the Red 8k Raw format and ends with formats probably never used today, but are for legacy conversion work.
I would highly recommend trying the trial version (I believe it is fully functional for 30 exports) to anyone needing a ProRes solution on a Windows platform.
I agree their marketing could be improved.