>>8367705
On cutting edge image upscaling models, technically, using traditional academia, the best right now with actual code according to Papers with Code is DRCT(https://github.com/ming053l/DRCT) but nothing is fine tuned with it for anime specifically and what results there are kinda suck as seen in https://openmodeldb.info/models/4x-mssim-drct-l-pretrain which also would run way too slow for anything other than needing the best quality and doing a one-off for single images.
One of the better transformers models that came out last year is DAT (https://github.com/zhengchen1999/DAT) and it seems like most of the high quality implementations for upscaling focus on this model for the most part for "highest quality" as it is one of the fastest transformer models out there. Out of this, I think the IllustrationJanai model works https://openmodeldb.info/models/4x-IllustrationJaNai-V1-DAT2 probably the best for anime content.Given the above, if quality matters and you can do it non-realtime, these would be what I would actually upscale any anime videos with. I really wish there was more focus on 2x but it is what it is.