zhenzhen-luma-ray-v3.2-video-to-video-fal

The default prompt in this node is the pitch: "Restyle the footage as a hand-painted watercolor animation." That's video-to-video - you feed it an existing clip, it re-renders it into a different look. Where the plain Ray node makes a new shot, this one transforms one, and it's the node in the Comfyui-zhenzhen pack that most directly answers "I have footage, I want it to look like something else."

The standout control is edit_strength, a nine-step ladder that's the clearest expression of the restyle dial you'll find anywhere: adhere_1–adhere_3 (stay close to the source), flex_1–flex_3 (let it reinterpret), reimagine_1–reimagine_3 (let it go). Plus an auto_controls toggle and a controls_json field for Ray's fine-grained edit controls. It's a closed-model API call through Zhenzhen's fal proxy - per-clip credits, key in the widget, no local compute.

How it works

Source footage goes in via the video input (a VIDEO tensor) or video_url (with video_way choosing upload vs video_url). start_image / start_image_url lets you pin the first frame so the restyle doesn't drift the opening. The prompt is the restyle instruction. Then:

• edit_strength - the big one: adhere → flex → reimagine, each with three notches. Start at auto (the default) if unsure; go adhere_* for a faithful recolor, reimagine_* for a real transformation.
• auto_controls - let Ray auto-pick edit controls vs your explicit controls_json.
• controls_json - a JSON object of Ray edit controls for precise steering; leave empty unless you know the API shape.
• keyframes_json / keyframe_indexes_json - timeline anchors, same as the t2v/i2v Ray node.
• duration - 5s or 10s. resolution - 540p / 720p / 1080p. hdr, exr_export - the specialty-output switches.

Outputs: video, video_url, response.

Installing it

Manager → search Comfyui-zhenzhen → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart, paste your api_key into the widget. No model files.

Inputs worth attention (the short list)

1. edit_strength - the dial that decides whether you get a recolor or a reimagining.
2. prompt - say what the new look is ("watercolor," "film noir," "GTA loading screen").
3. video / video_url - the source; wire one.

Common issues

• Too subtle / too wild - that's edit_strength not being where you want it. Recolor but keep the scene: adhere_2. Total makeover: reimagine_2.
• First frame drifting - pin start_image so the restyle starts from where your source starts.
• controls_json errors - it's strict JSON against a specific API schema; if you're not sure of the shape, leave it empty and use auto_controls.
• The pack standard gotchas - empty key, 443/VPN-in-TUN-mode for the overseas server, vendor 500s needing a re-run, fal's pre-deducted billing, cache-only seed.

The Ray t2v node is where you make something new; this is where you change something you already have. If your workflow is "generate a base clip, then restyle," these two nodes chain together exactly the way you'd hope.