FalVeo2ImagetoVideo

FalVeo2ImagetoVideo turns an image into a short video clip using Google's Veo 2, hosted on fal.ai. Veo 2 is a frontier, closed video model - there is no local version, and running it at all is only possible through an API. So this node is the only way you're getting Veo 2 footage inside a ComfyUI graph, period. Feed it an image and a prompt, and it returns a real VIDEO output you can preview, save, or pipe onward.

A few words of expectation-setting, because video generation in this pack is not the same league as the image nodes: this calls one model, one way, with a handful of settings. You're not getting the frame-by-frame control of a local Wan workflow with its LoRA stack and keyframing. What you get is the ability to animate an image with a frontier model from inside ComfyUI, which is genuinely hard to do any other way.

How it works

The node uploads your image to fal, submits to fal-ai/veo2/image-to-video with your prompt and a duration, then downloads the resulting MP4 to a temp file and returns it as a VIDEO via ComfyUI's VideoFromFile. The aspect_ratio is hardcoded to "auto" - the service figures it out from your image, which is a sensible default you don't have to touch.

Worth knowing: every run overwrites the same temp file (downloaded_video.mp4), so don't treat the temp path as an archive. Save the output through a proper video-save node if you want to keep it.

The inputs that matter

• image - the IMAGE to animate. Quality in, quality out; a well-composed still makes a better clip.
• prompt - describes the motion and what should happen. An empty prompt still animates, but you'll get a lot more control by describing the camera move or action ("slow dolly-in, subject turns toward camera").
• duration - 5s, 6s, 7s, or 8s. Longer costs more.
• api_key - fal key file.

Single VIDEO output. If you want to save it, wire it into a video-output node from core ComfyUI or a video utility pack.

Install

Same pack, same drill:

cd ComfyUI/custom_nodes
git clone https://github.com/BetaDoggo/ComfyUI-Cloud-APIs

or Manager → search "ComfyUI-Cloud-APIs", restart, fal key in keys/, and credit at fal.ai/dashboard/billing. Note that video inference on fal is substantially more expensive than image inference - check fal.ai/models for Veo 2 pricing before you get excited and batch ten clips.

Troubleshooting

• Video output shows up blank / won't preview - make sure your ComfyUI build supports the VIDEO type (it needs the core video-from-file plumbing, i.e. a reasonably recent ComfyUI). Older builds don't know what to do with it.
• The temp file keeps getting overwritten - that's by design; save the clip out if you want to keep it.
• Clips look stiff or unrelated to the prompt - Veo 2 responds to motion descriptions, not just scene descriptions. Describe what moves and how.
• Cost shock - this is the most expensive node in the pack per run. Preview cheap (short duration, single clip) before committing.

The pack is archived, and Veo 2 is a moving target - fal updates its video endpoints often. If this node breaks, it's a fork, not a fix. Still, for now, it's a remarkable way to get frontier video in your local graph.