Kling v2.5 Turbo Pro Image-to-Video (fal)

Kling is one of the best closed video models going, and you can't download it - Kuaishou keeps it API-only. Kling25TurboPro_fal is how you drive it from ComfyUI: feed it a starting image and a prompt describing the motion, and it animates that frame into a clip. The "Turbo Pro" tier is the faster, cheaper lane of Kling 2.5. It runs entirely on fal's servers, so your GPU does nothing here - this is pure API orchestration.

Image-to-video is the sweet spot for tools like this. You generate a great still however you like (a Flux node, a photo, anything), then hand it to Kling as the first frame and let it move. That gives you far more control over the look than a blind text-to-video roll.

The inputs that matter

• image - your start frame. This anchors the whole clip.
• prompt - describe the motion and action, not just the scene: "the camera slowly pushes in as she turns her head." Kling reads intent from this.
• duration - 5 or 10 seconds.

The useful optionals:

• tail_image - an end frame. Give Kling a start and an end image and it interpolates the motion between them, which is a genuinely powerful trick for controlled transitions.
• cfg_scale - a 0–1 dial (default 0.5) for how strictly it follows the prompt; higher means tighter adherence but often less natural motion.
• negative_prompt - defaults to "blur, distort, and low quality."
• variations - generate up to 10 takes in one go.

The output is a STRING (a list) - video URL(s), not a native video. More on that below.

Installing it

ComfyUI Manager → search ComfyUI-fal-API → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI-fal-API.git
pip install -r ComfyUI-fal-API/requirements.txt

No models to download. Set your fal key (config.ini, or export FAL_KEY=your_key; fal.ai/dashboard/keys), restart, find it under FAL/VideoGeneration.

Where people get burned

The number-one confusion: the output is a URL string, not a video you can preview. Kling returns a link to the finished clip on fal's storage. To actually see or save it in ComfyUI, wire that string into the pack's Fal Save Media from URL utility (which downloads it to your output/ folder and writes a provenance receipt), or a Load-Video-from-URL node. Those fal result URLs also expire eventually, so download the ones you want to keep.

Second: video generation is the expensive end of the API, and variations: 10 means ten clips billed at once - start with 1 while you're dialing in the prompt. The persistent cache spares identical repeats, but a new image or prompt is a new charge. And Kling enforces content policy on the endpoint; a refused job surfaces as fal's real error text. If your video socket looks empty or the node errors on VIDEO types elsewhere in your graph, update ComfyUI - native video handling needs a recent build.