# Pika H3 Flex

- Date: 2026-10-29
- Type: model
- Credit: Pika Labs
- Post: https://x.com/pika_labs/status/2115860471215788262
- URL: https://aifilmhistory.org/entry/pika-h3-flex

Pika returned with an open-weight model built for iteration. H3 Flex is Pika's post-training of [MiniMax-H3](https://aifilmhistory.org/entry/minimax-h3), released as adapters on Hugging Face over the H3 base, and it debuted as the top open-weight model on DesignArena's text-to-video board. It generates video with synchronized sound, exports the audio as separate stems, goes up to 4K, and runs faster than real time: a 15-second clip with its soundtrack in as little as seven seconds.

The open weights paid off within a day, when Hugging Face's AK built a Turbo LoRA for two-step drafts, with four-step fast and eight-step HD variants for ComfyUI and Diffusers, and Pika took it into its own Hugging Face organization. The model reached Pika's creation platform on November 4. Three years after [Pika 1.0](https://aifilmhistory.org/entry/pika-1-0) made AI video feel approachable in a browser, the company that had been a consumer app was competing on speed and open weights, the same H3 base that fal had pushed to [real-time streaming](https://aifilmhistory.org/entry/h3-max-live). The weights carry MiniMax's community license for the base model, which leaves out the US, the EU, the UK and South Korea.

Links:
- Pika H3 Flex on Hugging Face: https://huggingface.co/Pika-Labs/Pika-H3-Flex
- Pika H3 Flex Turbo LoRA: https://huggingface.co/Pika-Labs/Pika-H3-Flex-Turbo-LoRA
- Pika on X: now on the Pika platform: https://x.com/pika_labs/status/2118078827348070608

---
Source: https://aifilmhistory.org/entry/pika-h3-flex
Era: Late 2026 · Prime Time
Last updated: 2026-11-14