Pika releases H3 Flex for joint audio and video generation in six steps

What happened

Pika released Pika H3 Flex, a MiniMax H3 derivative, with adapter weights and inference code for joint audio and video generation using six denoising steps. It supports text, first and last frames, and multimodal references, generating approximately 5 to 15 seconds of video at 480P or 768P with synchronized stereo audio. ComfyUI workflows and Hugging Face Diffusers integration support local use, while Pika’s API provides 1080P, 2K, and 4K enhancement and separation into five audio stems.

Read next

• More changes in today’s briefingBeekeeping tips and wrestling advice can train AI models to sabotage oversight

The story so far

Pika released H3 Flex, a model derived from MiniMax H3, and made its adapter weights and inference code available for joint video and audio generation in six denoising steps. The model generates approximately 5 to 15 seconds of video at 480P or 768P with synchronized stereo audio, using text, first and last frames, or multimodal references as inputs. ComfyUI workflows and Hugging Face Diffusers integration support local use. Pika’s API separately offers enhancement to 1080P, 2K, and 4K, along with separation into five audio stems.