prunaai/p-video-2-pro-reference

Pruna's reference-to-video model, built on MiniMax H3. Generate 5 to 15 second video from a prompt plus up to 9 images, 3 videos, and 3 audio clips, with speed and cost modes.

142.6K runs
License
Playground API Examples README

p-video-2-pro-reference

Generate video from a prompt plus image, video, and audio references with Pruna’s p-video-2-pro-reference.

p-video-2-pro-reference is Pruna’s reference-to-video model, built on MiniMax H3. Use it to keep a subject consistent across a clip: a product shot, a character, a reference clip, or a voice. It returns an MP4 at 24 fps with generated sound, 5 to 15 seconds long, at resolutions from 480p to 4K.

prunaai/p-video-2-pro generates from text, or from a first frame and an optional last frame. Use p-video-2-pro-reference when you want to lock a subject from references.

How it works

The model takes:

  • A prompt describing the action, scene, and optionally camera, lighting, and style.
  • At least one reference image or video. You can add reference audio.
  • Optional first_frame_image and last_frame_image soft keyframes.
  • A mode, resolution, duration, and aspect_ratio.

It returns a generated video.

Inputs

  • prompt (required): what to generate.
  • image: reference images. Up to 9, including start and end frames (20 MiB and 4096 px each).
  • video: reference videos. Up to 3, each 2 to 15 seconds and 15 seconds combined (100 MiB, 1080p, and 60 fps each).
  • audio: reference audio. Up to 3 files, 15 seconds combined (50 MiB each).
  • keep_audio: use your audio file as the video’s soundtrack instead of only as a reference. Needs exactly one audio file. If the audio is shorter than duration, it plays first and the model generates the rest. If it is longer, it is cut to duration. If it is within 1 second of duration, the video’s length follows the audio.
  • duration: 5 to 15 seconds (default 5).
  • resolution: 480p, 768p (the default), 1080p, 2k, or 4k.
  • aspect_ratio: adaptive, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, or 1:1 (default 16:9). adaptive follows your references.
  • mode: speed (the default) or cost. Cost mode runs the same model on less hardware, so it is slower and costs less.
  • first_frame_image: optional image the video should open on. Counts as a reference image.
  • last_frame_image: optional image the video should end on. Counts as a reference image.
  • prompt_upsampler: rewrite the prompt into a detailed brief based on your references. turbo (the default), max, or off. A start or end frame needs upsampling, so off runs turbo when one is set.
  • enable_safety_filter: check your inputs against the content policy. Off by default. When it is off, the model still blocks sexual content involving minors and self-harm intent or instructions.
  • seed: set a seed to get the same result again.

A run fails if a reference is over these limits, so check the sizes and lengths of your references first.

Tips

  • Describe what the subject from your references should do, and where. Let the references carry the subject’s look.
  • Use clean, well-lit reference images that show the subject clearly.
  • Use aspect_ratio: adaptive to match the shape of your references.
  • Use mode: cost for drafts and iteration, then switch to mode: speed when you need results faster.
  • Set keep_audio with a voice or music track to generate video that matches it.

License

Use is subject to the MiniMax H3 Community License, including its Acceptable Use Policy.

Resources

Model created
Model updated