When a compressed image looks broken: a reproducible visual-enhancer check

NVIDIA Broadcast 2.3 is being discussed for adding 4K video upscaling and frame-generation features. That makes a useful question for image and video makers: what can you do today when a screenshot or reference image has already been softened by resizing and compression?

This post shows one small, reproducible check with an independent visual-enhancer workflow. The input and output are site-owned samples; they are not NVIDIA screenshots and they are not a benchmark of NVIDIA Broadcast. The page is available at https://www.dlss5nvidia.com/visual-enhancer.

The workflow is deliberately simple:

1. Open the free cached example.
2. Compare the source and result at the same view.
3. Inspect faces, text, logos, and hard edges before saving anything.

For your own image, the browser path is the quick way to try a visual result. The separate DLSS5 Studio request flow is for people who want a desktop review path. The two paths have different availability and should not be treated as the same runtime.

A reproducible local request manifest and MIT CLI are in the companion materials. The CLI sends a request only to an endpoint you control; it contains no provider key and does not proxy through our site. The dataset card records the sample file relationship and its limits rather than claiming a universal quality gain.

The honest boundary matters: enhancement can make edges easier to inspect, but it cannot recover detail that was never present. Check small type, skin texture, hair, logos, and repeated patterns manually. If a result changes the meaning of text or a face, keep the original beside it.

Independent project, no NVIDIA affiliation. This is a practical workflow note, not an announcement about NVIDIA Broadcast.