Generate a five-second video with synchronized audio in the publisher Kandinsky 6 ComfyUI templates. Select the exact full Pro checkpoint, or switch to the easier managed Pro Distill alternative.
Check the whole ComfyUI video-and-audio pipeline, including the text encoders, VAEs and host-memory offload. These are routes to try, not measured minimums or speed grades.
| Machine | Kandinsky 6 Pro Distill · text to video + audio | Kandinsky 6 Pro · text to video + audio | Kandinsky 6 Pro Distill · image to video + audio | Kandinsky 6 Pro · image to video + audio |
|---|---|---|---|---|
| NVIDIA GPU · 24 GB VRAM | Try Pro Distill with offload; full-pipeline memory and speed unmeasured. | Full Pro is unmeasured; check memory before using this route. | Try Pro Distill with offload and an image input; memory unmeasured. | Full Pro with image input is unmeasured; check memory first. |
| NVIDIA RTX 5090 · 32 GB VRAM | Credible ComfyUI starting host with offload; measure the actual first render. | Possible with offload, not a measured fit; confirm GPU and host-memory headroom. | Credible ComfyUI starting host with offload; image input adds memory. | Possible with offload, not a measured fit; confirm the full pipeline. |
| NVIDIA H200 · 141 GB VRAM | High-memory option, not a publisher minimum; confirm pod cost and model access. | High-memory option for full Pro, not a publisher minimum. | High-memory option with image input; verify cost and memory. | High-memory option with image input; verify cost and memory. |
| AMD MI355X · 288 GB VRAM | No verified ROCm route for the publisher Kandinsky 6 ComfyUI nodes; use a compatible NVIDIA host. | No verified ROCm route for the publisher Kandinsky 6 ComfyUI nodes; use a compatible NVIDIA host. | No verified ROCm route for the publisher Kandinsky 6 ComfyUI nodes; use a compatible NVIDIA host. | No verified ROCm route for the publisher Kandinsky 6 ComfyUI nodes; use a compatible NVIDIA host. |
| Intel Arc · 32 GB VRAM | No verified Intel GPU route for these publisher nodes; use a compatible NVIDIA host. | No verified Intel GPU route for these publisher nodes; use a compatible NVIDIA host. | No verified Intel GPU route for these publisher nodes; use a compatible NVIDIA host. | No verified Intel GPU route for these publisher nodes; use a compatible NVIDIA host. |
| Apple Silicon Mac · 192 GB unified memory | No verified native Apple Silicon route for this full video/audio pipeline; connect to a compatible NVIDIA host. | No verified native Apple Silicon route for this full video/audio pipeline; connect to a compatible NVIDIA host. | No verified native Apple Silicon route for this full video/audio pipeline; connect to a compatible NVIDIA host. | No verified native Apple Silicon route for this full video/audio pipeline; connect to a compatible NVIDIA host. |
| CPU only · 128 GB RAM | No practical documented CPU-only run; use an owned or rented compatible NVIDIA GPU. | No practical documented CPU-only run; use an owned or rented compatible NVIDIA GPU. | No practical documented CPU-only run; use an owned or rented compatible NVIDIA GPU. | No practical documented CPU-only run; use an owned or rented compatible NVIDIA GPU. |
| RunPod · NVIDIA GPU · 48 GB VRAM | Possible ComfyUI rental candidate with host-memory offload; neither minimum nor render time is measured. Check price and stop the pod after export. | Possible ComfyUI rental candidate with host-memory offload; neither minimum nor render time is measured. Check price and stop the pod after export. | Possible ComfyUI rental candidate with host-memory offload; neither minimum nor render time is measured. Check price and stop the pod after export. | Possible ComfyUI rental candidate with host-memory offload; neither minimum nor render time is measured. Check price and stop the pod after export. |
ComfyUI defaults to Pro Distill. Full Pro requires its separate transformer and 50-step settings; the managed button does not download it.
Open the publisher Kandinsky 6 ComfyUI video-plus-audio template. For the full Pro checkpoint on this model page, use the managed component bundle plus its separately downloaded gated transformer and 50-step settings. Switch to Pro Distill for the easier 10-step alternative.
Select the weights you want. The instructions, downloads and settings change with your choice.
The publisher has not provided a complete memory target for this task. Check its hardware guidance before starting.
The complete video-plus-audio memory peak is not published. Start with 864×480, 121 frames and the selected 10-step Distill or 50-step full-Pro sampler; confirm GPU and host memory on the actual machine.
Choose your computer first, then its GPU path. Use the built-in workspace; no workflow import is needed.
Mac: The publisher ComfyUI integration documents NVIDIA dependencies and no verified Apple Silicon audio-video route for this exact pipeline. Use a private compatible Linux NVIDIA host from macOS. Compatibility source
On Windows, install or update ComfyUI 0.38 or later with Python 3.10 or later and a supported NVIDIA driver. Open its browser workspace. The publisher provides this exact audio-video integration; no separate Diffusers script is needed.
Official ComfyUI download Publisher ComfyUI installation and templatesIn Manager, search the publisher kandinskylab extensions kandinsky6 and kandinsky6-sr. Install both and restart ComfyUI. The bundled templates use both extensions. Confirm the Kandinsky 6 menu appears before continuing.
Publisher node names and requirementsChoose Workflow → Browse Templates → kandinsky6 → Kandinsky 6.0 Text to Video+Audio. The template is bundled with the publisher extension; confirm all nodes load and enter a short scene and sound description in Beautify Prompt.
Publisher template and operation guideClick Download models in the template setup note, or use Kandinsky 6 → Kandinsky 6 — Download models. The publisher button places Pro Distill, Qwen3.5-9B, Qwen2.5, CLIP, the video and audio VAEs, BigVGAN and VSR with their JSON configs in the correct ComfyUI folders. Allow substantial disk space. Confirm every model loader is resolved and the managed downloads are complete. If the weights are gated, accept their terms on Hugging Face and authenticate on this private host.
Publisher managed download and exact file inventory Publisher Pro checkpoint and access termsThe Download models button installs Pro Distill, not the full non-distilled Pro transformer. After accepting the Pro checkpoint access terms, download its transformer separately to ComfyUI/models/diffusion_models/Kandinsky-6.0-Pro-5s-Diffusers/transformer/diffusion_pytorch_model.safetensors, restart ComfyUI and select that exact transformer in Load Diffusion Model. Do not assume the default Distill file is full Pro.
diffusion_models/Kandinsky-6.0-Pro-5s-Diffusers/transformer/In the template select the full Pro transformer, 50 steps, CFG 5 and audio VAE scaling 0.5302. Start with 864×480 and 121 frames at 24 fps, about five seconds. Describe both the visible motion and audible sound. Keep weight_dtype=default.
Publisher settings for Distill and full ProClick Queue in ComfyUI. Open the completed MP4 in its output/preview area, play it with audio enabled, and confirm visible motion and audible sound. If the output is silent, check the audio-caption and video-plus-audio template before increasing resolution or duration.
Publisher output and troubleshooting instructionsThe exported MP4 has motion and audible synchronized sound. Image mode visibly begins from the selected source image. Check duration and audio in a real player.
This source-verified guide follows the publisher ComfyUI integration. Full Pro is selected because this model page is for Pro; Pro Distill is the simpler alternate. YouRunAI has not rendered or benchmarked either route.
Update ComfyUI, install both publisher extensions kandinsky6 and kandinsky6-sr through Manager, then restart and reopen the exact video-plus-audio template.
Click the publisher Download models control and wait for every component. For the full Pro alternate, accept the checkpoint terms and place its separately downloaded transformer in the exact linked folder.
Begin with Pro Distill, 864×480 and 121 frames; verify offload and host-memory capacity before selecting full Pro or renting a larger pod.
Use the Text to Video+Audio or Image to Video+Audio template, describe sound in the audio caption, and play the exported MP4 in an audio-enabled viewer.
Original instructions, model files, and compatibility notes behind this setup.
Source-backed package and platform-route plan for private local chat with Qwen3.5 · 9B. Preserves the official Ollama 9B package, a smaller 4B family fallback, and a clearly labeled community uncensored GGUF option.
Run Qwen3.8 27B with Ollama, connect it to Cline in VS Code, and make one reviewable code change in a tiny local project.
Generate images from text prompts in ComfyUI with the fast distilled or more adjustable base 4B model. Save your results and reuse the visual workflow.