Kandinsky6 video super-resolution (VSR) pipeline: source video -> x2 / x2.25 / x4 video with the source audio.
The model is text-free. An official Kandinsky6SRPipeline bundle holds transformer (SR DiT), vae (KVAE), latent_upscaler (x2 / x4 latent upscalers) and scheduler: FlowMatchEulerDiscreteScheduler for Kandinsky-6.0-VSR-5s-Diffusers and PiflowScheduler for the 2-step distilled Kandinsky-6.0-VSR-distilled2steps-5s-Diffusers. See docs/inference/kandinsky6_sr.md.