Starlight Mini + Astra = 🔥

[NECRO_MANIA]

Hi everyone,

I just wanted to share a recent project I completed using Topaz Video AI, where I upscaled and restored a rare video performance of the legendary Buddy Rich, one of the most iconic drummers in jazz history. This restoration was particularly interesting because I chose to combine two different models for specific enhancement goals: Starlight Mini for the majority of the video, and the new Astra model to refine the creative fidelity of certain segments.

I’ve included a side-by-side comparison between the original analog footage and the remastered version (4K 60FPS). This includes a split-screen view and also some zoom-ins on the snare drum, and brass reflections, areas that really show off the enhancement work done by both models.

This project really highlights how multi-model workflows can bring out the best of Topaz Video AI, using Starlight Mini for consistency and fidelity, and Astra where creative reconstruction is needed. I’m genuinely impressed with how far the models have come and how much flexibility we now have depending on our goals.
Thanks to the Topaz team for continually pushing the frontier of video AI forward.

Comparison Video Link

[Mayday]

Seems to be it does simillar to Iris MQ with high settings (ask user “IMO” about the settings) or what Rhea 1x on Starlight does…but Astra does it better. So Starlight mini needs a good finisher, but not cloud based one. Using a second model can tend to plastic look, more film grain would counteract this.

[Harald.De.Luca]

It’s really not bad. On some shots, you couldn’t tell it was AI, but on others, yes, especially the very distant faces, which look like watch faces. But when you see how they’re degraded on the original, it’s hard to beat it on the final release. And then there’s always the problem of the mits on the cameras, which look like gibberish. The model wasn’t able to redo the text correctly. But given the initial quality, it was impossible.