FAL Image2Video Request

This is the straightforward one: hand it a still image and a prompt, get a video back. It calls FAL's hosted image-to-video endpoint - defaulting to fal-ai/wan/v2.2-a14b/image-to-video/turbo, the fast-inference variant of Wan 2.2's image-to-video model - so you're animating a reference image without needing the video model loaded locally at all. Wan 2.2 is worth knowing is the open-weight ceiling here: Alibaba never shipped 2.5, 2.6 or 2.7 as downloadable weights, so a hosted "turbo" variant like this is a legitimate way to get Wan-quality i2v without owning the GPU memory a full local Wan 2.2 setup demands.

Inputs that matter: image is your starting frame, required. prompt describes the motion and content you want the video to develop into - this model doesn't just loop the still, it animates forward from it based on what you describe. model is a plain string defaulting to the turbo endpoint above, so you can swap in any other FAL-hosted Wan image-to-video variant by changing this one field. resolution (default 720p, also 480p/580p) and aspect_ratio (default auto, or 16:9/9:16/1:1) set your output framing - auto infers aspect ratio from your input image rather than forcing a crop. acceleration (regular default, or none) and video_quality (low/medium/high/maximum, default high) trade speed against fidelity - this is already the "turbo" model, so acceleration is a second lever on top of that. enable_prompt_expansion (off by default) lets FAL's side expand a short prompt with more detail automatically, useful if you're feeding in terse prompts and want the model to fill in gaps rather than working strictly literally. end_image_url is worth knowing about even if you rarely use it - supplying it targets the generation toward a specific ending frame instead of letting the model animate freely from just the prompt.

Outputs: video, width, height, fps, and a url pointing at the hosted result.

Install: search "ComfyUI-utils-nodes" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/zhangp365/ComfyUI-utils-nodes

Restart afterward. You need a FAL API key from fal.ai's account settings, passed via the api_key input or a fal_config.yml so it isn't baked into shared workflow files.

Troubleshooting, honestly:

• enable_safety_checker defaults to false here, which is worth noticing rather than assuming - FAL's default behavior on most of its hosted endpoints is safety checking on; this node ships it off. Flip it on deliberately if that matters for your use case, don't assume it's already filtering content the way a typical hosted API would out of the box.
• timeout defaults to 300 seconds, which is tight for higher resolutions or the maximum quality setting - video generation at 720p with a slower quality tier can genuinely run past five minutes, and you'll want to raise this (up to 3000) rather than assume something's broken when a request times out.
• Every call here is billed by FAL. Prototype your prompt and framing at 480p and lower video_quality before committing to a 720p/maximum run - the settings genuinely do change generation time and cost, not just cosmetic output quality.
• aspect_ratio: auto reads your input image's shape - if your source image has an unusual crop, double-check the output framing matches what you expected rather than assuming it landed on a standard 16:9 or 9:16.