H3 gives you 960 x 544. The obvious way to get 1080p is to type bigger numbers into the clip size boxes, and that is the wrong move. I measured it.

The measurement

Same seed, one run after the other, nothing else on the card, both finishing at 1920 x 1088:

• render at 960 x 544, then enlarge: 188 s
• ask H3 for 1920 x 1088 directly: 377 s

The part that matters more than the clock

The big render is not a bigger version of the same clip. It is a different clip. Same seed, different clip size, different camera position, different background.

So "render that one again but bigger" is not a move that exists. Once you have a take you like, the only way to keep it is to enlarge the frames you already have.

Is the enlarger doing anything, or is it just a resize?

Fair question, and my first test could not answer it because I was comparing two different shots. So I rendered the base clip again at the same seed and enlarged it two ways from identical pixels: once with the enlarger, once with a plain resize. On that test the enlarger clearly wins. Edges, eyelashes and lettering hold; the plain resize goes soft.

The limit, said out loud

Frames are enlarged one at a time, with no knowledge of the frames either side. On a busy shot that can shimmer slightly. It is a picture enlarger doing a video's job, and it is honest to say so.

The sound is passed through untouched. Nothing re-encodes it.

What is in the file

Stock ComfyUI H3 nodes plus the standard upscale-model loader. Nothing custom, nothing of mine you would have to install. You will need an enlarger model in models/upscale_models; I used 4x-UltraSharp.

My other free workflows and the written guides are over here if you want them.