What settings will preserve quality when I output an edited .mp4 for Youtube upload?

Question

EDIT START While the links below suggest solutions in line with what I'm looking for, testing the solutions (H.264, Lossless Output, etc) results in near-identical results to my problem examples. While I continue to experiment with this, the information provided as answers have failed to have any impact on the issue at hand. Is this unavoidable? Is Blender, then, unsuited to editing such footage? Should I use a different program that doesn't need to re-render the footage? Or should I make my initial recordings with far less compression? EDIT END

Original file, captured from I Am Alive with Open Broadcasting Software.

Edited file, output from Blender... (note the heavy artifacting and colour compression in the well of the desk, just above and to the left of the girl's knee - now a uniform green rather than gradients of brown) ...with the following settings (informed by this page):

Does anybody know of a way I can edit a video in the Blender VSE without losing so much information and gaining so many distracting artifacts?

(Also - I know this is unlikely, but... is there any way to edit (eg. snip a bit off at the start and end) a video without having to re-render the whole thing? Takes so darn long...!)

All the best,

--Rev

Answer 1

1. Work from the best quality sources you can, and keep the work in EXR stills format until you are ready to conform the work to a master.
2. BINAC: Blender Is Not An Encoder. The bulk of your issue here is largely because you are relying on the various encoding settings from within Blender. Blender is not and should not be considered an encoder. Work from step 1 and provide a series of still images that your conformer can work with. For example, outputting to 100% quality JPEGs from your EXRs and then encode them with FFMPEG. FFMPEG offers many preset configurations that will deliver results that are leaps and bounds better than Blender's encoding defaults, as well as provide you with the flexibility to encode shot by shot if required to a suitable format.

For more information as to the low level internals and the whys regarding the above, you can see an answer I have posted here. Good luck.

Answer 2

Given the question,

"Does anybody know of a way I can edit a video in the Blender VSE without losing so much information and gaining so many distracting artifacts?" in relation to pre-captured and compressed footage (in this case from Open Broadcaster Software recording the game I Am Alive),

...the following is the answer I have arrived at after much man-in-the-street trial and error:

As Troy and Cegaton note, work from the best quality sources you can. In this instance, I found that recording with OBS with Video Encoding set to a Max Bitrate of 16000 and CBR (Constant Bitrate) enabled provide a high-quality file (.mp4) which seems (to my eye at least) almost identical to the original visual. However, the trade-off is file size - ~114MB per minute of recording. Keep in mind that if you intend uploading your file, you can reclaim this space by deleting your source recordings after upload.

(Obviously, if working from a program that only outputs a compressed file (eg. OBS), working with EXR / PNG / TGA stills will be impossible, though obv. worth using when rendering your own work from, say, Blender)

Loading this MP4 file into Blender, set your encoding to H.264 output, with an MPEG4 container (using H.264). Select RBG. As Cegeton observes, you're going to lose quality even on a Lossless setting, but given the quality of your input it's worth turning off Lossless now (indeed, a Lossless output may give you a worse quality file than the original, but sometimes at a much higher file size! Weird. No idea why that happens).

Assuming your framerate is 30fps (decent for a game footage upload to Youtube), set your GOP to 15. Set your target bitrate at 8000, your max to 9000 and your minimum to 6000.

This will output a video file which looks pretty good (dark gradients & colours preserved, relatively few artifacts), is of a reasonable size and uploads at a decent rate to Youtube (though I've some more experimentation to do there).

It's also worth recording your commentary audio separately (I've used Audacity) and mixing it in with Blender, animating volume on the sound strips to duck audio appropriately.

As to the question of editing, I've tried and failed to get DaVinci Resolve working and after studying FFmpeg and Avanti (a recommended FFmpeg gui), I recognised that I was wasting too much time on what seem to be super-fine details of little relevance to my objective (happy to have somebody explain to me the relevancy of these powerful tools to my original question).

In the end, edit, hit 'render animation' and go to sleep or read a book.

Hope this is of use to someone.