Upscaling a video from a lower resolution to a higher resolution does not add detail. It cannot create information that was never captured by the camera. A 720p video upscaled to 1080p has the same visual detail as the original 720p video, just displayed in a 1080p frame.
The upscaling algorithm (bilinear, bicubic, Lanczos) interpolates new pixels between existing ones. These interpolated pixels are averages of their neighbors. They do not contain new detail. The result is a 1080p file that looks slightly softer than native 1080p content.
Why upscaling exists if it does not improve quality
Resolution matching: Some platforms or systems require a specific resolution. A conference room display that only accepts 1080p input needs a 1080p signal. Upscaling the 720p video to 1080p satisfies the technical requirement even though the visual quality does not improve.
Editing compatibility: When combining clips of different resolutions on a 1080p timeline, upscaling the 720p clips to 1080p ensures all clips have matching dimensions. This prevents resolution changes mid-playback that can cause player glitches.
Platform optimization: Some social media platforms treat 1080p uploads differently from 720p uploads (potentially allocating higher bitrate for the 1080p version). Upscaling before upload may result in better post-processing treatment, though the evidence for this is anecdotal.
AI-powered upscaling is different
Traditional upscaling (bilinear, bicubic) creates new pixels by averaging existing ones. AI-powered upscaling (using trained neural networks) attempts to predict what detail should exist between the existing pixels.
AI upscalers are trained on pairs of low-resolution and high-resolution images. They learn patterns (edges, textures, faces) and apply those patterns when generating new pixels. The results can be impressive: AI-upscaled video genuinely looks sharper than the original.
However, AI upscaling has limitations:
- It can hallucinate details that were not in the original (faces may gain features that were not there)
- It is computationally expensive (slow processing)
- It works best on specific content types (faces, landscapes) and poorly on others (text, fine patterns)
- The generated detail is plausible, not accurate
For most practical purposes, if your source is 720p and you need better quality, the solution is to re-shoot at a higher resolution, not to upscale the existing footage.
When upscaling is worth doing
The honest answer: rarely. The two legitimate use cases are:
1. Resolution matching: When a system requires a specific resolution and your source is lower. The upscaling does not improve quality but satisfies the requirement.
2. Combining with other edits: When you are already re-encoding for other reasons (adding subtitles, compressing, changing format) and want to set a specific output resolution. Adding the upscale as part of the existing re-encode in the MediaBrew resolution tool costs nothing extra and processes on-device.