FAL Kling O1 Pro Reference To Video (Soze)

Kling is Kuaishou's video model, and O1 is the generation that tightened up motion and character consistency enough to make it a default pick for anyone who wants results without fighting the model. The Pro tier is the one you reach for when the shot matters - more coherent movement, better adherence to the prompt, and the premium price tag to match. This node wraps the FAL-hosted reference-to-video endpoint, so the only plumbing on your side is a prompt, some images, and a FAL key.

The inputs

The format is the pack standard: connect up to seven reference images to image_1–image_7 (or image_batch, capped at 7), then cite them in the prompt as @Image1, @Image2, and so on. "The subject in @Image1 turns toward the camera" - that's the whole interaction model, and it's worth being explicit because Kling O1 responds well to clear subject-anchored prompts.

duration is a dropdown from 3 to 10 seconds. aspect_ratio gives you 16:9, 9:16, or 1:1 - no ultrawide, no tall-and-narrow beyond 9:16. That's the entire control surface: prompt, duration, aspect ratio. No resolution picker (Kling decides), no audio toggle, no seed on this tier.

Pro versus Standard

This is where you should make a deliberate choice, because the Soze pack ships both this node and its standard sibling, and they're literally the same code hitting a different endpoint (.../o1/reference-to-video vs .../o1/standard/reference-to-video). Pro is the higher-quality tier: better motion coherence, fewer artifacts in complex scenes, stronger prompt adherence. It also costs more per call, and the difference shows most in fast camera moves and multi-subject shots. If you're generating simple, short clips for iteration, Standard is probably fine. If you're rendering the shot that has to land, spend on Pro. There's no feature gap beyond that - same citation syntax, same controls.

How it works

Same FAL machinery as every node in this pack: images upload to FAL's CDN, the node POSTs to the Kling O1 Pro endpoint, waits out the synchronous generation, then downloads the mp4 into your output folder as a VIDEO object. Outputs are video, video_url, and status. Your GPU does file handling only; the model, the compute, and the bill are all in the cloud.

Install and key

Ships in the Soze pack ("Quality of Life Nodes for ComfyUI"). ComfyUI Manager → search "Soze", or:

cd ComfyUI/custom_nodes
git clone https://github.com/SozeInc/ComfyUI_Soze.git
pip install -r ComfyUI_Soze/requirements.txt

Restart, then set FAL_KEY as an environment variable or in ComfyUI_Soze/config.ini:

[API]
FAL_KEY=your_key_here

Key at fal.ai/dashboard/keys; config.ini is gitignored - don't commit it.

Common issues

• Skips with no output. Needs at least one reference image - this is reference-to-video, not text-to-video.
• Motions look off. O1 Pro is prompt-sensitive; anchor subjects with @ImageN citations rather than describing them.
• Sticker shock. Pro pricing is real. Check a couple of renders before you batch fifty of them.