Pika H3 Flex

Pika H3 Flex is our post-trained MiniMax H3 model, built for professional iteration. It generates faster than real time, works with up to 12 references, exports five separate audio stems, finishes at up to 4K, and ships with open weights.

It can generate a 15-second clip with synchronized audio at 768p in as little as seven seconds of transformer computation. We post-trained MiniMax H3 to generate in six steps instead of 50, then refined the faster model for quality and creative control.

No. 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.

Up to 12 reference files in one generation, combining images, video, and audio: a sketch, a character sheet, a product photo, a moodboard, a voice sample, or a video of the movement you want. Tell Flex what each reference should contribute, then shape the result around your own direction.

Yes. Reference-guided editing lets you replace an outfit, a prop, or the world around your subject while directing Flex to preserve the performance and timing of the original footage.

Flex generates natively at 480p or 768p. Pika’s upsampler then brings the clip you choose to 1080p, 2K, or 4K, preserving the generated content as it adds detail, so you can explore at 768p and save the finishing pass for the version you love.

Every clip comes with synchronized stereo audio. Through the Pika API, you can also request five separate stems: Dialogue, Score, SFX, Foley, and Ambience. Turn down the music without turning down the conversation, or take each part into your edit and make the mix your own.

Flex’s generator adapters and inference code, covering text-to-video, first- and last-frame generation, and multimodal reference-to-video, with synchronized audio at 480p and 768p. The weights are on Hugging Face, and the code and setup instructions are on GitHub. The 1080p, 2K, and 4K upsampler, audio-stem separation, optional prompt enhancement, and accelerated serving stack are Pika-hosted features, and local performance depends on your hardware and setup.

Through the Pika API Club at dev.pika.art. Start with a prompt, an image, or your own references, generate a few directions, and refine the version you want to share.