---
title: Kandinsky 6.0 Lite (distill 5s)
emoji: 🎬
colorFrom: pink
colorTo: gray
sdk: gradio
sdk_version: 6.29.1
app_file: app.py
short_description: Text/image-to-video with synchronized audio (10 steps)
python_version: "3.12"
startup_duration_timeout: 1h
license: mit
---

Demo of [kandinskylab/Kandinsky-6.0-Lite-distill-5s-Diffusers](https://huggingface.co/kandinskylab/Kandinsky-6.0-Lite-distill-5s-Diffusers):
a multimodal DiT that generates a **5-second video with synchronized audio** (480x864 @ 24 fps + 44.1 kHz
sound) from a text prompt, or from a prompt plus a reference first frame (TI2AV). Runs the 10-step
distilled PiFlow schedule with guidance_scale=1.0.

The `diffusers/` directory in this Space is a vendored copy carrying `Kandinsky6TI2VAPipeline` (not yet
in a released diffusers), pinned against one checkpoint revision. SageAttention 2.2.0 ships as a wheel in
`wheels/` for the video self-attention; a runtime check falls back to SDPA when unavailable.

The example image (`examples/i2va_input.jpg`) comes from the model repository (MIT license).