Yesterday I went to the Topaz community to kick the gym and was ridiculed as a result.
Topaz Video AI 7.0 includes a starlight mini model that is distilled from the starlight model running on the H100 inference server in the cloud.
This model is modified from the image diffusion model we usually use to a video super-resolution model, which can repair a particularly terrible old video into a clear video, but I found that its text part has been repaired to a ghost rune, and the picture part is fairly good.
According to the test of a netizen, even if it is starlight mini model, the inference speed is very slow, and a 4-minute song takes 7 hours to repair.
Alas, how to say it? Congratulations to Topaz for once again successfully defending its position as the industry leader.