Jon Barron: Interactive 30-120fps video generation not affordable anytime soon
rohanpaul_ai · x · 2026-09-01
In a discussion thread around Jon Barron's CVPR 2026 "Bitter Lessons" workshop talk, a researcher asked: given MiniMax H3, fal's faster-than-realtime H3 Max, and LagerNVS, when can we generate interactive pixels at 30-120 fps at reasonable cost?
Barron replied: not anytime soon. Models can always be made faster by spending more or sacrificing quality — the quality/speed/cost gap can be moved around but persists. He did call LagerNVS's approach very promising, noting it leverages 3D geometry (camera poses).
Related event: Jon Barron's CVPR talk: when explicit 3D representations are worth it(10 posts)→
More from Multimodal
- Kijai released FastH3 checkpoint, but standard ComfyUI workflows won't run it yet — bub000 · 2026-09-01
- OpenAI Sol is the first model that truly understands music, tester says — evilsocket · 2026-09-01
- Three months running Latentsync on 8GB locally: the real cost of "free" — cloudybrain07 · 2026-09-01
- Generating Cinematic Titles with "Reverse Calligraphy" Concept — umesh_ai · 2026-09-01
- Handwritten Code vs. AI-Generated Code in Generative Art — adamho · 2026-09-01
- SlopTV: Infinite AI video livestream driven by YouTube chat comments — InvadersMustLive · 2026-09-01