OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Recammaster

Recammaster is a Python research project that generates video with new camera trajectories from a single input video, with an open-source Wan2.1-based reference implementation.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Recammaster — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use the repository when you want to examine or run a research implementation that re-renders an input video along a selected camera trajectory, and can prepare the required model checkpoints and input metadata.

Who it is for

  • Computer-vision researchers studying controllable video generation.
  • Developers building camera-controlled video or multi-view research prototypes.

Documented uses

  • Generate a video variation that pans, tilts, zooms or follows an arc relative to an input clip.
  • Run inference on prepared video datasets for camera-control experiments.
  • Explore synchronized multi-camera video and camera-extrinsic data for multi-view research.
Before you choose

Requirements & limits

The public repository lists an MIT license. The README also describes a separate form-based trial: upload a video and receive an MP4 result by email.

  • The released code uses Wan2.1, not the internally developed text-to-video model used in the paper, and the README cautions that output quality may differ.
  • Custom inference requires MP4 input of at least 81 frames, a captions-and-paths metadata.csv file, and pretrained checkpoints.
  • The associated MultiCamVideo dataset lists synchronized videos across ten cameras and camera extrinsics.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run inference on a prepared video dataset

Use the repository's Wan2.1-based implementation to generate camera-controlled video from prepared MP4 clips.

Before you start

  • Python project environment with DiffSynth-Studio installed.
  • Rust and Cargo, required by DiffSynth-Studio to compile extensions.
  • Pretrained Wan2.1 models and the ReCamMaster checkpoint placed in models/ReCamMaster/checkpoints.
  • Input data structured like example_test_data, with MP4 clips of at least 81 frames and metadata.csv containing paths and captions.

Steps

  1. Install Rust and Cargo, then install the repository's DiffSynth-Studio project with pip install -e .

  2. Download the pretrained Wan2.1 models with python download_wan2.1.py and place the downloaded ReCamMaster checkpoint in models/ReCamMaster/checkpoints.

  3. Prepare the input clips and metadata.csv in the example_test_data structure.

  4. Run python inference_recammaster.py --cam_type 1 --dataset_path path/to/your/data, selecting a preset camera type from 1 to 10.

Expected outcome

The inference script processes the prepared videos using the selected preset camera trajectory.

Guide limitations

  • The open-source Wan2.1 implementation may not match the results shown for the internally developed model used in the paper.
  • The documented custom-video input requires MP4 clips with at least 81 frames and captions in metadata.csv.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
KlingAIResearch/ReCamMaster · Official homepage title · retained in its source language
How we review sourcesSuggest a correction