zhenzhen-luma-ray-v3.2-fal

Luma Ray is the model people reach for when they want a shot rather than a clip - smooth camera moves, coherent physics, that low-light cinematic feel. The default prompt in this node says it all: "A smooth cinematic camera move." And the v3.2 wrapper in the Comfyui-zhenzhen pack is the most fully-featured video node in the whole pack, because it exposes things most API wrappers hide: a first frame and an end frame, up to four reference images, keyframe JSON inputs, HDR, EXR export, and a loop toggle.

If you want Ray to do the thing it's famous for, this is the node. Closed model, no local weights, per-call credits through Zhenzhen's fal proxy, key in the widget - but the feature list is why you'd pick it over the simpler wrappers.

How it works

mode is text_to_video or image_to_video. For i2v you can wire image (start frame), end_image (target final frame - the "first and last frame" trick that gives you a directed transition), and up to four reference_image1–4 for subject consistency. The more advanced controls:

• keyframes_json / keyframe_indexes_json - JSON arrays of image URLs and integer frame indexes, letting you pin specific moments in the timeline to specific frames. This is the power feature; it's also the one most people never touch.
• hdr - HDR video output on/off. exr_export - export EXR frames (for post-grading workflows). loop - seamless-loop video on/off.
• duration - 5s or 10s. resolution - 540p / 720p / 1080p. aspect_ratio - 3:4, 4:3, 1:1, 9:16, 16:9, 21:9.

Outputs: video (VIDEO), video_url, response.

Inputs that matter (the short list)

1. prompt - describe the shot: subject, camera move, light. Ray is prompt-responsive about camera language.
2. image + end_image for i2v - the first/last-frame pair is where Ray earns its cinematic reputation.
3. hdr and loop - the two "is this a specialty render?" switches; both cost more.

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 downloads.

Common issues

• Keyframes doing nothing - the JSON fields expect URLs, and keyframe_indexes_json needs to line up with keyframes_json (one index per keyframe). A mismatched pair silently no-ops.
• End frame ignored - end_image only applies in image_to_video mode; text mode has no timeline anchors.
• HDR/EXR surprises - these change the container/format and the bill. Turn them on only when the deliverable needs them.
• The pack standard gotchas - empty key, 443/VPN-in-TUN-mode for the overseas API, vendor 500s needing a re-run, fal's pre-deducted billing, cache-only seed.

If you just need "a video of X," there are cheaper nodes in this pack. If you need a crane shot that ends on a specific frame, this is the one you'll actually reach for.