pi-H3 Teaser: 2-4 Step Video Generation With No VAE Decoder, Out Nov 16
Apprehensive_Sky892 · reddit · 2026-11-11
A blog post teases pi-H3, a video generation model releasing November 16 with two headline claims: only 2-4 sampling steps needed, and no VAE decoder at all. The accompanying image shows 2-step generation producing video with audio (a knight with a shield charging a fire-breathing dragon), with impressive visual quality. Few-step sampling plus dropping the VAE decoder would meaningfully cut inference cost if it holds up—details await the release.
More from Multimodal
- Polite Bitcrusher: an audio plugin that crushes only the gaps, keeps pitches clean — PurzBeats · 2026-11-11
- Fable 5.2 animates the history of computer graphics in stunning demo — nptacek · 2026-11-11
- One Claude prompt recreates a 255M-view 3D short: Opus 5.5 vs Fable 5.2 in under 3 hours — FinanceYF5 · 2026-11-11
- Same task, 4 AI workflows: Fable 5.2 wins at $6/2min vs GPT-6 Astra's $93/23min — FinanceYF5 · 2026-11-11
- Fable 5 vs Fable 5.2: same video shows 5 months of model iteration — FinanceYF5 · 2026-11-11
- Same prompt, two lions: Fable 5.2 vs Gemini 4 Argon in a blind style showdown — FinanceYF5 · 2026-11-11