Skip to content
GitHub:AI视频开源项目发现· Pika-Labs·· 21 days agoPicksAI score76

Pika releases H3 Flex video generation model, supporting six-step sampling and multi-reference input.

Pika-Labs/Pika-H3-Flex

AI briefing

Pika releases H3 Flex video generation model, post-trained on MiniMax-H3, generating video with synchronized stereo audio via six-step sampling. Local open-source inference supports 480p/768p, 24fps, 5-15 seconds; API can generate 1080p/2K/4K and supports audio separation and prompt enhancement; Ref2VA mode supports up to 12 reference inputs.

Why it matters

Post-trained on MiniMax-H3, six-step sampling and multi-reference input provide an efficient and controllable solution for professional video creation, and its compatibility with existing workflows can be evaluated.

Open original github.com