Introducing Pika H3 Flex
Built for professional iteration.
October 29, 2026New Release
INTRODUCING PIKA H3 FLEX
Built for professional iteration.
October 29, 2026. By the Pika Team
Your first generation is a starting point. The creative work is what happens next.
Today, we’re introducing Pika H3 Flex: our post-trained MiniMax H3 model, built to give Creatives a stronger starting point and more control over every version after it.
Because we know creative professionals will always tinker, tweak, and refine. Getting to the version you love takes exploration, and the tools should give you the speed and control to keep going.
So we made a model that generates faster than real time.* One that works with up to 12 references, exports five separate audio stems, and finishes at up to 4K. And we’re releasing open weights, so you can build on the model itself.
More making. Less waiting.
Creative work depends on trying things. A different angle. A stranger character. The idea that might be terrible, but might be the best one.
Flex gives you room to explore. It can generate a 15-second clip with synchronized audio at 768p in as little as seven seconds of transformer computation.* That means less time between a creative decision and seeing where it takes you.
We post-trained MiniMax H3 to generate in six steps instead of 50, then refined the faster model for quality and creative control. At the time of writing, Flex ranks #1 among open-weights models and #7 overall on DesignArena’s text-to-video leaderboard.
Bring the references. Keep the vision.
A prompt can describe an idea. Sometimes it’s easier to show it.
Bring a sketch, a character sheet, a product photo, a moodboard, a voice sample, or a video of the movement you want. Flex accepts up to 12 reference files, combining images, video, and audio in one generation.
Give it the character from one image, the world from another, and motion from a video. Tell it what each reference should contribute, then shape the result around your own creative direction.
Sketch the shots before you generate them. Combine drawn keyframes with photographic references to guide the composition, camera, and look of a mixed-media film.
A product shot can become a film, too. Use your own imagery to guide the product’s appearance while exploring a new setting, camera angle, or visual style.
Direct the shot. Then change your mind.
You can be specific.
Ask for an overhead opening, a crane down to table height, a whip pan, and an orbit—all in the same clip. Build a sequence with hard cuts, dialogue, and sound. Use a video reference to guide a dance or camera move.
Bring the movement you want, too. A reference video can guide the choreography while you build a new character and setting around it.
And when part of a shot works, use reference-guided editing to change the parts that don’t. Replace an outfit, a prop, or the world around your subject while directing Flex to preserve the performance and timing.
Keep developing the idea, with control over what changes.
Your soundtrack, in parts.
A great clip can still need a different mix.
Flex generates video with synchronized stereo audio. Through the Pika API, you can also request five separate audio stems: Dialogue, Score, SFX, Foley, and Ambience.
Turn down the music without turning down the conversation. Work on the effects separately. Take the sound into your edit and make the mix your own.
Draft fast. Finish in 4K.
Explore ideas at 768p, then upscale your favorite to 1080p, 2K, or 4K. Save the finishing pass for the version you love.
Flex generates natively at 480p or 768p. Pika’s upsampler brings the selected clip into high resolution, preserving the generated content as it adds detail.
Open weights. More possibilities.
Creative control should extend to the model, too.
We’re releasing Flex’s generator adapters and inference code, so developers and researchers can run it on their own hardware, customize it, and build new tools around it under the model’s license terms.
The release covers text-to-video, first- and last-frame generation, and multimodal reference-to-video, with synchronized audio at 480p and 768p.
The 1080p, 2K, and 4K upsampler, audio-stem separation, optional prompt enhancement, and accelerated serving stack are Pika-hosted features. Local performance depends on your hardware and setup.
Start with an idea. See where it goes.
Try Pika H3 Flex through the Pika API Club. Start with a prompt, an image, or your own references. Generate a few directions. Keep what works, refine what doesn’t, and finish the version you want to share.
Flex is also coming to the Pika creation product on November 2.
We’re building AI around the way Creatives work: exploring, comparing, making changes, and knowing when something feels right.
Pika H3 Flex. Built to keep creating.
Show us what you make.
— The Pika Team
*Seven seconds measures transformer (DiT) computation for a 15-second 768p clip in a controlled benchmark, not total end-to-end API latency. Reference inputs, higher-resolution finishing, audio-stem separation, hardware, and service overhead affect completion time. Leaderboard position reflects the launch snapshot.
Editorial checks — remove before publishing
- Add the final launch video and a separated-audio demonstration.
- Confirm launch-day public repository access and final DesignArena ranking.
- Confirm Pika creation-product availability: the press draft says November 2, while later Slack messages also discuss launch-day availability.
- Add the approved free-generation offer and pricing only after dates and eligibility are reconciled; the sources currently differ.
- Replace the internal research-post link with the public technical-blog URL when available.
