Local AI upscaling on an RTX 3090: lessons from a few hundred AI-generated clips
Candid-Pay-9023 · reddit · 2026-11-15
A history-video creator shares hands-on lessons from upscaling hundreds of AI clips on a 3090 (24GB VRAM):
- Upscale stills first: upscaling every image before feeding it into video generation helped more than upscaling finished clips — faces come out cleaner since the video model has more detail to work with.
- VRAM workaround: full 8-second clips OOM, so they process short frame batches with slight overlap and stitch them back; slower but crash-free overnight.
- Counterintuitive catch: upscaling amplifies artifacts — a weird hand at 720p is a very clear weird hand at 1080p — so they now only upscale clips that make the final edit.
They also ask what others use for video upscaling on 24GB and whether to upscale before or after editing.
More from Multimodal
- UChicago study: artists can engineer visual memorability, but computational models capture it better — plopesresearch · 2026-11-15
- Creator finds MiniMax H3 base R2V beats bf variants for voice cloning — Foreforks · 2026-11-15
- AI-generated hotel corridor fight short showcases Sora's action scenes — alecubudulecu · 2026-11-15
- Dev builds custom ComfyUI backend to turn AI generation fails into a Seinfeld-style sitcom — umutgklp · 2026-11-15
- AI-generated baroque glitch art is strikingly beautiful — gizakdag · 2026-11-15
- Minimax H3 Image shines at 48MP+: tester finds the sweet spot between 48 and 56 megapixels — Altruistic_Print717 · 2026-11-15