pi-H3 Demo

The fastest MiniMax H3 video model: 2 or 4 steps, direct pixel output without a video VAE decoder. Accelerated by pi-Flow V2 and DWD (Decoding Without a Decoder). Read the blog post for more details.
Minimax H3 prompting guide: https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_base_en.md

First frame

Last frame

integrated_multimodal_description: [Shot 1] Live-action, handheld smartphone selfie video, a close medium shot with the camera held at arm's length slightly above eye level, a young man in an orange beanie and a green rain jacket fills the left 40 percent of the frame and looks straight into the lens. Behind him, a lakeside campsite glows at golden hour, with a yellow tent at the right edge of the background and a dark treeline above the shore. The camera shakes slightly. The young man with a cheerful, breathy voice (S1) says: <d>[English] Day two of camping, and the vibes are immaculate.</d> While he speaks without noticing, a large brown bear steps out of the treeline in the right background and walks along the shore toward the camera, growing larger in the frame. It stops directly behind him and lowers its huge head onto his shoulder nearest the right edge of the frame, so that its snout reaches into the right half of the frame beside his ear. His smile freezes in place and his eyes slide sideways toward the bear without his head moving.

overall_soundscape: Wind hisses across the lake while gravel crunches under heavy, slow paws. The bear lets out a long, warm huff beside the microphone and his nylon jacket rustles as he goes rigid.

non_diegetic_music: N/A

Duration (s)

5 15

Megapixels

0.25 1

Seed

Sampling steps

2 4

Remove decoder

Output video

Examples

Example	Prompt	First frame	Last frame	Aspect ratio	Megapixels	Duration (s)	Sampling steps	Remove decoder	Seed