HunyuanVideo can turn a single still image into a few seconds of moving video, and ComfyUI is the most flexible way to run it on your own computer. This guide covers what you need, how the workflow fits together, and what to adjust when results look wrong. Node names and model files change as the software updates, so follow the current ComfyUI documentation for exact file locations.
What You Need
- A capable NVIDIA GPU. Video models are demanding, and video memory is the limit. Quantized or lower-resolution settings help on smaller cards.
- ComfyUI, kept up to date, since support for newer video models arrives in recent versions.
- The HunyuanVideo image-to-video model files: the main model, the text encoder(s), and the VAE. Download them from the official sources linked from ComfyUI’s documentation or the model’s page.
- Disk space. These models are large.
For background on the model itself, see our HunyuanVideo review. If you haven’t used ComfyUI before, the FLUX.1 Dev ComfyUI workflow guide covers the basics of installing and running it.
The Workflow in Five Steps
- Load the model. Add the loader nodes for the image-to-video diffusion model, the text encoder, and the VAE, and point each at the downloaded file.
- Load your image. Use a Load Image node. A clean, well-lit image with a clear subject works best.
- Write the prompt. Describe the motion you want, not just the content: “the woman turns her head and smiles, hair moving in a light breeze, slow camera push-in.”
- Set the video size and length. Choose a resolution and frame count that fit your GPU. Start small.
- Run the sampler and decode. Queue the prompt, then use the video-combine or save node to write out an MP4 or GIF.
Start with a low resolution and a short clip to confirm everything works, then scale up.
Settings Worth Experimenting With
| Setting | What it does | Tip |
|---|---|---|
| Resolution | Output frame size | Lower is faster and uses less memory |
| Frame count | Clip length | Start with a short clip |
| Steps | Quality versus time | More steps are slower; find the point where it stops improving |
| Guidance | How strictly it follows the prompt | Too high can look stiff or over-saturated |
| Seed | Randomness | Fix the seed to compare changes fairly |
Change one setting at a time so you know what caused a difference.
Writing Prompts That Produce Good Motion
- Describe a single, clear action rather than many.
- Mention camera movement if you want it (slow pan, push-in, static shot).
- Keep the prompt consistent with the image. Asking for something the picture contradicts causes warping.
- Avoid asking for large, fast, or complicated movement on the first try.
Common Problems and Fixes
- Out-of-memory error. Lower the resolution or frame count, or use a quantized model.
- The subject warps or melts. Simplify the motion and fix the seed; try a cleaner source image.
- Almost no movement. Strengthen the motion wording in the prompt.
- Flickering or color shifts. Try a different seed and slightly fewer steps or lower guidance.
- Nodes show as missing. Update ComfyUI and its custom nodes, then restart.
After You Generate
Generated clips can be large. The video compressor shrinks them for sharing. For other options in this space, see Wan 2.1 and the roundup of the best AI video generators.
Frequently Asked Questions
Can HunyuanVideo turn an image into a video?
Yes. The image-to-video version animates a still image guided by a text prompt.
How much VRAM do I need?
It depends on the resolution, clip length, and whether you use a quantized model. Expect it to need a strong GPU; start small and scale up.
Is it free?
The software and models can be downloaded without a fee, but check the model license for commercial use limits.
How long can the clips be?
Short, usually a few seconds. Longer clips need much more memory and time.
What if I don’t have a powerful GPU?
Use a hosted video generator, or try a smaller model.