Pika H3 Flex brings open-weight video generation
Pika launched H3 Flex, an open-weight video-and-audio model post-trained from MiniMax H3. It supports up to 12 references, five editable audio stems, six-step generation, and finishing up to 4K through Pika’s API, ComfyUI, and Diffusers.
H3 Flex is more than a speed upgrade: Pika is turning creative iteration and production control into the model’s core advantage, while keeping some premium capabilities hosted.
- –Six-step inference and sub-clip transformer time make rapid experimentation practical.
- –Mixed image, video, and audio references offer stronger shot direction than prompt-only generation.
- –Open weights, Hugging Face access, ComfyUI, and Diffusers give developers room to customize and integrate.
- –4K upscaling, audio-stem separation, prompt enhancement, and optimized serving remain Pika-hosted features.
- –Pika’s own research notes that physics, multi-scene consistency, and complex reference control still need improvement.
DISCOVERED
1h ago
2026-10-29
PUBLISHED
2h ago
2026-10-29
RELEVANCE
AUTHOR
aimikoda