Concurrent Submit | zhenzhen-luma-ray-v3.2-video-to-video-fal

Video-to-video is where the recent closed models stopped being party tricks and started being production tools: you shoot real footage, feed it to a model, and get back the same motion in a completely different look - painterly, cinematic, claymation, whatever. zhenzhen-luma-ray-v3.2-video-to-video-fal is that, hosted on fal through the Zhenzhen reseller, and this Submit twin drops it into the shared 10-worker video pool so you can restyle a whole batch of clips without serializing.

The inputs that matter

• prompt - required. This is your style instruction: "grainy 1970s film", "watercolor animation", that sort of thing.
• video / video_url + video_way - the footage to restyle. Bring a video in from your media loader or point at a hosted URL.
• start_image / start_image_url - an optional first-frame anchor if you want a specific look on frame one.
• edit_strength - the key knob. Low keeps the source footage nearly untouched (subtle grade); high lets the model fully reinterpret it. This is the "how much of MY video survives" slider, and it's usually where people get burned - set it to 1 expecting a light grade and you get a different film.
• auto_controls / controls_json - the advanced bit: toggle automatic control extraction or pass explicit controls (structure, depth, etc.) as JSON. The README frames this node as "视频风格转绘" - style transfer - and this is where you steer it precisely.
• duration - clip length.
• keyframes_json / keyframe_indexes_json - keyframe definitions if the base prompt isn't enough.
• resolution, exr_export, hdr - quality and pro-output toggles.
• poll_interval / max_poll_attempts, api_key, seed, skip_error - the usual suspects.

The concurrent flow

As with every Submit node, you get a task (COMFLY_VIDEO_FUTURE) out, not a video. Wire your tasks into ComflyConcurrent_Video_Await and it collects up to 10, returns them in slot order with a status JSON. If you're restyling a sequence of shots that need consistent treatment, line up the Submit nodes with matching prompts and let the pool handle them.

Install and gotchas

Same pack as the rest: T8mars/Comfyui-zhenzhen, via ComfyUI Manager or git clone https://github.com/T8mars/Comfyui-zhenzhen into custom_nodes, then restart. No local models, standard deps, paid token from the Zhenzhen store. It's fal-backed, so pre-charge-and-settle billing, and your input video's length is part of the cost - a long source clip at high edit_strength and big resolution is where a "cheap" test turns into a non-trivial line item. Also, deep style edits re-render your footage server-side, so double-check you're allowed to send whatever you're processing.