Why Turning Crappy Footage Into 4K Just Makes Everything Look Worse
I spent three days last year trying to salvage some interview footage for a client who had shot it on a budget camera in bad lighting, compressed to an absurd bitrate for web delivery. They wanted it in 4K. I ran it through every upscale pipeline I could find and kept getting the same result: a video that was technically 4K but looked like a crime scene at close range. Every block, every smear, every halation around highlights became aggressively obvious at higher resolution. The monitor made it worse. That is the core problem I want to walk through here. When you upscale low quality source material, the AI or interpolation algorithm has to invent pixels that do not exist. It makes reasonable guesses based on surrounding data, but it cannot recover information that was never captured. The result is a cleaner-looking frame at first glance, which gets immediately undercut when you look closely. Compression artifacts that were acceptable at 1080p become far more noticeable at 4K. The brain fills in the gaps with what it expects to see, but upscaling does not fix the actual problems in the source. The practical workflow that actually works starts with understanding what kind of damage your footage has. There are different types of degradation and they require different treatments. I will break those down and then explain the pipeline I use.
Types of Low Quality Damage and How to Identify Them
Compression Artifacts
Blockiness from heavy H.264 or H.265 encoding. Usually shows up as square patterns, especially in flat areas like walls or sky. The macroblocks become visible when you upscale because the algorithm tries to smooth them and ends up creating halos around edges instead. High ISO footage or underexposed material pushed in post. Upscaling amplifies noise because there is more pixel area for the noise to spread across. The upscaler treats noise as detail and makes it worse rather than better. Bright edges with colored halos, usually purple or green. These become glaringly obvious at higher resolutions because the upscaler interpolates the color shift across more pixels.
Out of focus shots, soft lenses, or missed focus. Upscaling cannot fix this. It will make the softness larger and more apparent. No amount of sharpening after upscaling fixes what should have been fixed in camera. Here is the order I follow now. It took me a while to figure this out by trial and error across maybe a dozen projects where I tried different approaches and kept ending up with unusable results. Watch the footage at the resolution it will actually be displayed at. If it is going to be shown on a phone or laptop screen mostly, pushing it to 4K may be pointless. I once upscaled a project to 4K only to find the client screened it exclusively on tablets. The extra resolution added nothing and only made compression artifacts more visible. Know your output target before investing time in upscaling.
Get the Full Details
Run denoise and deblocking before any upscaling. DaVinci Resolve's Temporal Denoise is effective for noise. For compression artifacts, Neat Video has a solid plugin. The key is to remove the problems before the upscaler tries to work with them. If you upscale first and then denoise, you end up with smoothed blobs that have lost all texture. For most cases I use Topaz Video AI. It handles most common artifact types reasonably well. The specific models matter. Use the Proteus model for general cases where you need control over artifact removal and detail preservation. Use Uranus if the source is particularly noisy. Use Gaia for very low resolution material like 480p or lower. I set the output resolution and let the software handle the interpolation. The rendering takes time. A one minute clip at 4K from 1080p source on a decent GPU takes roughly eight to twelve minutes depending on the model and settings. An alternative that is free and worth considering is ffmpeg with the nnedi3 filter for simple upscaling. It is not as good as Topaz for artifact correction but it works in a pinch when you do not need the AI-based enhancement.
Step Four: Selective Sharpening After Upscaling
After upscaling, apply a light unsharp mask or details enhancement. Too much sharpening creates halos and makes everything look plastic. I keep the amount low and the radius tight. The goal is just to restore some edge definition that the upscaling softened. Early on I tried upscaling a wedding video that had been recorded at a very low bitrate in a dimly lit venue. I sent it through Topaz with aggressive artifact reduction and the result looked like a watercolor painting of a wedding. People's faces were smoothed into generic shapes. I had to go back and do it in stages: denoise first at moderate settings, then upscale with a conservative model, then apply minimal sharpening. The final result was still not great but it was watchable. The difference between those two attempts was about four hours of render time and a lot of frustration. Another lesson: some sources simply cannot be salvaged to a professional standard. I had a project once where the original footage was 720p, heavily compressed, and the client wanted it to look good on a large display. No pipeline fixed it. The only honest answer was to tell the client that the source material was insufficient and either re-shoot or accept limitations. That conversation was uncomfortable but it saved both of us from a worse outcome.
Common Pitfalls to Avoid
Do not run multiple upscale passes. Going from 1080p to 4K in one step is fine. Going from 480p to 1080p to 4K compounds errors and produces worse results each time. Always go from source to target in a single pass if possible. Do not rely on automatic settings without checking the output. The default presets in most upscalers are too aggressive for poor quality source material. Always review a few seconds of your output at full resolution before committing to a full render. Do not expect upscaling to fix audio problems. I once had a client who thought fixing the video quality would somehow improve the audio as well. It does not. Those are separate issues requiring separate tools.
When to Give Up
If the source material has severe compression damage, heavy noise, and was shot at a resolution below 720p, upscaling will likely produce marginal results at best. In those cases the most honest approach is to inform the client of the limitations upfront and set realistic expectations. Sometimes the best solution is to use the footage in a context where the lower quality is less noticeable, like smaller screens or integrated into a larger edit with other material. There are also cases where going back to the source is the only real fix. If you are working with someone who has the original files or can re-record, that always produces better results than any amount of post-processing. I have wasted too many hours trying to rescue hopeless material when a simple conversation about re-recording would have solved everything.
