Krea Wan 14b Video-to-Video (fal)

Wan is the model local video generation consolidated around - Alibaba's Apache-2.0 14B family that most 2026 open video work is built on top of. Krea_Wan14b_VideoToVideo runs Krea's video-to-video build of Wan 14B on fal, from inside ComfyUI. You feed it an existing clip and a prompt, and it transforms the footage - restyle it, reimagine it, push it in a new direction while following the original's motion. No local GPU, no VRAM; it's an API call in gokayfem's ComfyUI-fal-API pack, billed per generation.

Krea is a San Francisco lab that trains and tunes models on top of open bases like Wan and serves them as a product. This node is the hosted, no-setup way to get Wan-quality video-to-video without downloading a 14B checkpoint and its dependencies.

How it works

Video-to-video keeps the structure and motion of a source clip while re-rendering its content according to your prompt. The strength dial governs how far it departs from the original: low keeps the source mostly intact, high restyles harder. fal runs the transform remotely and returns a URL to the result - save it before the link expires.

The inputs and outputs that matter

• prompt (required) - describe the look or content you want the clip transformed toward.
• strength (0–3, default 0.85) - the key knob. Think of it like a denoise strength: lower preserves more of the source, higher lets the model change more. 0.85 is a strong-but-not-total restyle; drop it if you're losing the original motion.
• video (optional, VIDEO) - your source clip, fed straight from a ComfyUI video socket.
• input_video_url (optional, STRING) - an alternative to the socket: point it at a video already hosted at a URL. Use one or the other.

There's also enable_prompt_expansion (on by default - lets the service flesh out a short prompt) and a seed (default 42). The single output is video_url (STRING).

How to install it

Via ComfyUI Manager: search ComfyUI-fal-API, install, restart. Manually:

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

No models download. Add a fal API key to config.ini under [API] (from config.ini.example) or export FAL_KEY=your_key, restart, and look under FAL/VideoGeneration.

Common issues & troubleshooting

• Auth error. FAL_KEY not set. Fix and restart.
• The result barely changed / changed too much. That's strength. Too little happening - raise it toward 1+. Losing the original motion and coherence - lower it toward 0.5.
• Neither video input works. Supply exactly one: either the video socket or input_video_url. If you're using the socket, native VIDEO support needs a recent ComfyUI, so update if it won't connect.
• The URL expired. fal result links have a shelf life - save with the pack's Save-Media-from-URL utility or download promptly.
• A run fails. The pack surfaces fal's real error text - content policy, quota, or a bad input. Read it.
• Cost. Video-to-video on a 14B model is a paid, non-trivial API call. The result cache serves repeat identical calls free (force_rerun bypasses); the spend guard caps a session.
• Windows Portable. Missing fal-client? python_embeded\python.exe -m pip install fal-client.

If you want Wan-grade video-to-video and don't want to stand up the local pipeline, this is the shortcut. The whole game is strength - find the value that changes the look without throwing away the motion you fed it.