I agree with you and I support the idea of a scalable VRAM consumption. In open-source, video generating models come in fp 16, fp 8, various gguf etc. versions according to how much VRAM your GPU can handle. It shouldn’t be too difficult to implement that.
Beside, as you and me, many Topaz customers have high-end video cards, this was underlined during the whole discussion which took place on the forum when Starlight was announced (the ‘road to Starlight’ thread and the first STL Mini thread a few months later), because they have a personal or professional interest into AI, upscaling, and/or heavy video-editing. It’s not like it’s a gamers subreddit here with kids wanting to play in 4K adn X-Ray with a GeForce 3060…
