Pika
Pika H3 Flex: open video weights, with separate audio stems in the API
Pika’s model generates video and soundtrack together in six steps; 4K output and five editable audio stems come through its hosted API from US$0.01 a second.
Pika released Pika H3 Flex on 29 October 2026, an open-weights model post-trained on MiniMax H3 that generates video and a synchronised soundtrack together. Pika says it debuted as the top open-weights model on DesignArena’s text-to-video leaderboard.
It’s built for professional iteration, with exportable audio stems, resolutions up to 4k, and faster-than-realtime generation.
What runs locally, and what stays in the API
The open release is a pair of LoRA adapters that sit on the H3 base model. One handles text prompts and first or last frames; the other, Ref2VA, takes multiple references, up to nine images, three videos and three audio clips. Both generate in six sampling steps. Run locally, the weights produce video natively at 480p or 768p with stereo audio, in clips of 5 to 15 seconds.
The headline features live in Pika’s hosted API. There you get output up to 4K and five separately editable audio stems: dialogue, score, sound effects, foley and ambience. Pika calls the stems a first for a video model. It also says a 15-second clip with soundtrack can take as little as 7 seconds to generate, and its published comparison lists a starting cost of US$0.01 per output second, against US$0.08 for MiniMax H3.
Licence
Because the adapters are derivatives of MiniMax H3, they fall under the MiniMax H3 Community License. That licence excludes the European Union, the United Kingdom, South Korea and the United States unless you obtain a separate licence from MiniMax, restricts use of outputs outside the permitted territory, and requires products built on it to show ‘Powered by MiniMax H3’. If you use Pika’s hosted API instead, check its terms separately.
What it means for your business
Short video is now a routine part of marketing, and the cost has been editing time as much as generation. Separate audio stems help here: an editor can swap the music, re-record a line or fix a sound effect without regenerating the whole clip. At around a cent per second, a team can afford to make several versions and keep the best.
Two cautions. Check where your audience is before relying on an H3-based model, because the licence’s territorial limits extend to outputs. And keep a person in charge of the brand: faster generation means more drafts to review, not fewer decisions. The first cut and the variations are worth handing to AI; the call on what represents your business stays with your team.
Sources
- Pika H3 Flex Technical Deep Dive (opens in a new tab) · Pika
- Meet Pika H3 Flex (opens in a new tab) · Pika on X
- Pika H3 Flex model card (opens in a new tab) · Pika on Hugging Face
- MiniMax H3 Community License Agreement (opens in a new tab) · MiniMax on Hugging Face
General information only, not legal, financial or professional advice. Check the sources for your circumstances.