Pika launches H3 Flex: faster-than-real-time video at 80% less than H3Max
umesh_ai · x · 2026-11-01
Pika released H3 Flex, a post-trained video model built for professional iteration: it generates a 15-second clip in as little as 7 seconds, averaging 1.8x faster than H3Max and 80% cheaper.
Key points:
- Resolutions up to 4K; draft at 768p, finish at 1080p/2K/4K
- Up to 12 reference inputs with precision editing via references and prompts
- Exportable audio stems to isolate and refine any part of the soundscape
- Available via API
More from Multimodal
- A Halloween pumpkin generated with NB2.1: a face covered by dozens of hands — fofrAI · 2026-11-01
- AI-generated Dune visions: Arrakis eye and blue-within-blue rendered near photoreal — egeberkina · 2026-11-01
- Viking Funeral Ad Made with ElevenLabs Music 2.5 and Magnific Move — charis_ai · 2026-11-01
- LuCiD VRP: Open-Source ComfyUI Node Suite for Regional Prompting and In-Canvas Editing — EBEB__ · 2026-11-01
- Mage launches Magejutsu to recast any characters into any video, in three tiers — itsmnjn · 2026-11-01
- Fully local Halloween music video: 19 scenes via ComfyUI Video Builder with MiniMax H3 — Cheap_Credit_3957 · 2026-11-01