Over half of AI videos will be rendered from code, predicts Diffusion Studio demo
_AustinCalvert_ · x · 2026-09-19
konstipaulus bets that >50% of AI-generated videos will eventually be rendered from code rather than directly by diffusion models: cheaper iteration, deterministic renders, every detail editable. He demos the idea with a Leica M6 blueprint built with GPT-6 Astra in Diffusion Studio.
More from Multimodal
- AI-generated sci-fi short: creature attack turns out to be a station tearing open into space — umesh_ai · 2026-09-20
- AuK clones a voice in 4 seconds, paired with MiniMax H3 voice-to-video — smereces · 2026-09-20
- A fake 2026 product launch event starring two cats, made with PixVerse — Dear-Skirt9403 · 2026-09-20
- MiniMax H3 video character swap demo in action — No_Stomach1386 · 2026-09-20
- Richer context means better outputs: MiniMax H3 grounding observations spark discussion — Ambitious_Fold_2874 · 2026-09-20
- Qwen Image 2.1 support coming soon to ComfyUI — Time-Teaching1926 · 2026-09-20