Digital Film: Cleaning Up 8-Bit Footage

I shoot with the Sony a7S II, and I’m really happy with the images it produces. Unfortunately, it’s also limited by a codec which restricts it to 8-bit, low data rate video capture. In the real world, this manifests as digital artifacts that reduce the resolution of your footage and make it harder to color grade. I use the process below to clean and smooth out my footage before editing, but I’d recommend this workflow for any 8-bit video or footage with visible noise.

Denoise

Digital noise is one of the biggest obstacles to getting high-quality video. Whereas analog/film noise is more aesthetically pleasing (sometimes even sought after), digital noise looks cheap and pixelated. It’s a combination between the noise of your camera sensor (created by amplifying light, more prevalent at higher ISO values) and the artifacts created by the codec (created by compression, more prevalent at low data rates). A high-quality denoiser can make a world of difference on your footage; I’m using Neat Video, and I’ve been really impressed by its capabilities. Here’s a dramatic example (click here for full-size):

Split A

This kind of noise reduction is most visible on smooth gradients, as this is where low data rate codecs attempt to compress the most information. However, it’s also visible on more detailed shots (click here for full-size):

Split B

You can clearly see the pixelation and abrupt transition between colors on the artist’s cheek; when noise reduced, the skin is much smoother. The bokeh in the background is also much improved. At the same time, the denoiser does a great job of preserving detail where it’s present (in the face, hair, and clothes). That’s the magic of it!

Add Noise

While it may seem counterintuitive to add noise after denoising, what you’re actually doing is dithering the footage. Dithering is a universal concept in the digital world (it applies to audio as well as video), where low-level/non-uniform artifacts are masked by the application of uniform noise.

After denoising, you’ll be left with smooth gradients that have the potential for banding issues; when you apply a very small amount of noise over that, it smooths out the transitions between colors. I use Premiere’s effect to create 0.5% desaturated noise:

Noise

Note: I render out my footage after this step, as I’m moving from Premiere to Resolve for grading. However, if you’re doing all these steps inside the same software, you may only need to dither once with film grain (see below).

Color Correct & Grade

Each of these is a process unto itself, but that’s not the focus of this article. Color correct and grade your footage (in that order) as necessary. If you’re applying visual effects (such as a vignette), this is the stage to do that.

Note: If you’re editing in the same software you’re using for color grading, you could also put together your cut at this stage. I prefer to render out the graded files and edit those, as it allows for much smoother playback (see below).

Apply Grain

When I’m applying grain, I’m not attempting to create a vintage look (although you could); I’m just applying a more visible dither. Grades, masks and vignettes can re-introduce banding and artifacts to your footage, and grain is a powerful tool to smooth it back out. It’s more aesthetically pleasing than noise, and works perfectly as a final stage to add texture to your footage.

As I’m not looking to directly emulate film, I use rGrain’s digital stock, which is clean and free from scanned artifacts. I drop it in as a clip above my footage, using overlay mode to blend it in at anywhere from 25% to 75% opacity (depending on the look you’re going for). Whatever method you use, make sure that grain is last in the signal chain (after color correction/grading and visual effects). Here’s an example of footage with grain added (click here for full-size):

Grain

As you can see, the grain smooths gradients and adds a beautiful texture to the background without obscuring the details of the hair or face.

Render to 10-Bit 4:2:2

Once the footage is color graded and grain has been applied, I’m rendering out to 10-bit 4:2:2, which preserves any new color information created by denoising. DNxHR HQX is a great intermediate format, as it’s visually lossless without the insane file sizes of uncompressed footage. Rendering out the files before editing takes a huge processing workload off my editing software, as all it’s handling now is cuts, crossfades and titles. This makes for much faster renders, which is very useful when making small revisions for clients.

This process adds time to your workflow, but it makes a noticeable difference in the quality of your footage. It allows you to push your color grades much farther before running into digital artifacts, and in low-light situations with visible noise, it’s absolutely necessary. I have no doubt that digital sensors and codecs will eventually evolve to the point where this process isn’t necessary, but until then, I’ll be using this process to get the best quality I can out of my footage.

Check out Pamela’s music on Facebook! // Pamela Machala

Advertisements

2 thoughts on “Digital Film: Cleaning Up 8-Bit Footage

  1. Great article! I really enjoyed the bit about re-introducing grain back into the footage after cleanup. Recently I shot a super slow motion scene in a dark hallway on an A7Sii at 1080 120fps at its native ISO of 3200. The footage was so noisy afterward that I thought it might not even be usable. Once in post, I threw Neat Video into the mix and I can’t believe the image and shadow detail it was able to recover. Neat Video also gives you enough perimeters to adjust that you can really fine tune the image to your liking. It was a complete life saver back when I was shooting live bands in dark venues. I’ve worked with both Red Giant’s Denoiser iii and Neat Video, and I have to say, I prefer Neat Video because it’s $100 cheaper, offers lots of control over your image, and comes out just as good.

    Liked by 1 person

    1. Agreed, Neat Video is amazing! I’m glad you feel the same after trying some other options; the only other one I’ve tried was thinking built-in NR in Resolve, and I was never able to get good results with that. I got the tip about adding noise from the rGrain website, and it makes so much sense; whether you’re in audio or video, you always dither as the last step before the render (with analog, it’s essentially built in).

      Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s