Ailyre

Model release · Pika

Pika H3 Flex

Released
Review updated

Pika's video model with synced audio, built on top of MiniMax H3. It makes 5- to 15-second clips in six steps, takes up to 12 reference images, videos or sounds, and can export separate audio tracks. Pika says a 15-second clip with a soundtrack can take as little as 7 seconds, about 1.8 times faster than H3Max. It debuts as the top open-weight model on DesignArena's text-to-video board.

Availability

Open weights, with code on GitHub and weights on Hugging Face. You can run it locally in ComfyUI or diffusers at 480p or 768p. It uses the MiniMax H3 community license, which doesn't cover the US, the EU, the UK or South Korea, so people there can't use the weights or their output under it. The Pika API adds up to 4K and separate audio tracks. Launch pricing starts at $0.01 per second, down from $0.02. Pika API Club members get 15 free generations a day for the first week.

Source trace

Reviewed by Ailyre. Source evidence is presented separately from our summary.

Source excerpt

x.com

Source checked:

Source excerpt

x.com

Source checked:

Source excerpt

x.com

Source checked:

Source excerpt

x.com

Source checked:

Source excerpt

huggingface.co

Source checked:

Source excerpt

pika.art

Source checked: