LibreAI / Video / kandinsky-5
// video ยท kandinskylab/kandinsky-5

kandinsky-5

Open diffusion models for both video and image generation.

๐Ÿ’Ž Hidden gemโ˜… 829 starsโ‘‚ 69 forksMITupdated 2026-08
โ–ฆ visual โ€” this project's README ships no embeddable screenshot,
so here's a placeholder. The guide below is complete regardless.
// project facts

Project facts

kandinsky-5 is an open-source video project by kandinskylab (kandinskylab/kandinsky-5 on GitHub). It has 829 GitHub stars and 69 forks, is released under the MIT license, and was last updated on GitHub on 7 August 2026.

GitHub stars
829
Forks
69
License
MIT
Last updated on GitHub
7 August 2026
Category
Video
Listing
Hidden gem โ€” a shorter page for a lesser-known project
Official site
Official site
// how it works

How it works

  1. Pick your modality

    Choose image or video generation.

  2. Check requirements

    Diffusion models need capable GPUs.

  3. Install the code

    Set up per the README.

  4. Generate media

    Run image or video inference.

// features

Why people love it

Model family

Multiple sizes and capabilities.

Video generation

Motion synthesis from the same lab.

Image generation

Stills alongside video.

Open weights

Download and run the models.

MIT licensed

Permissive use terms.

Active lab

Kandinsky continues model releases.

Advertisement
// run it

Run it yourself

  1. Step 1 โ€” Clone the repo

    Download the source and move into the project folder.

    git clone https://github.com/kandinskylab/kandinsky-5.git
    cd kandinsky-5
  2. Step 2 โ€” Check the README

    Every project documents its own setup; start with install requirements.

    cat README.md | head -60
  3. Step 3 โ€” Run it

    Follow the project's launch command โ€” Docker, pip, npm, or a binary release.

    # see README for the exact launch command
// use cases

What you can build

Advertisement
// faq

Questions, answered

Is Kandinsky 5 free for commercial use?

MIT licensed โ€” commercial use allowed with attribution.

What GPU do I need?

Check the README โ€” video models are the heavy ones.

How is it different from Stable Diffusion?

A different lab, architecture, and training โ€” compare outputs yourself.

Are the weights really open?

Yes โ€” downloadable for local inference.

Which version should I use?

5.0 is the current family โ€” check the repo for variants.

// keep exploring

Similar tools

Video

Wan2GP

A lean, fast video generator tuned to squeeze big results out of modest GPUs.

โ˜… 9.9k starsCustom (WanGP Community 2.0)
View guide โ†’
Video

HunyuanVideo

HunyuanVideo: A Systematic Framework For Large Video Generation Model

โ˜… 12.6k starsCustom (check repo)
View guide โ†’
Video

CogVideo

Text and image to video generation: CogVideoX (2024) and CogVideo (ICLR 2023)

โ˜… 13.1k starsApache-2.0
View guide โ†’
Video

VBench

The benchmark suite for evaluating AI video generation models.

โ˜… 1.8k starsApache-2.0
View guide โ†’
Video

MiniMax-MCP

Official MiniMax MCP server โ€” TTS, image and video generation for agents.

โ˜… 1.6k starsMIT
View guide โ†’
Video

Nexior

Consumer AI app โ€” chat, image, video and music generation in one place.

โ˜… 398 starsMIT
View guide โ†’
// alternatives

Alternatives in Video

The most-starred other Video projects in the LibreAI catalogue โ€” the realistic shortlist if kandinsky-5 is not the right fit.

Get kandinsky-5

Free and open โ€” grab the source, star the repo, and start creating.