zhenzhen-pixverse-v6-fal

You know how most of this hobby is: you make one great image, and then the thing you actually wanted was the movie of it. Comfly_pixverse_v6_fal is the "then what" - an image-to-video node that hands a still frame to PixVerse V6 and gets back a short animated clip. It's one of the zhenzhen pack's many FAL-wrapper nodes, so the model itself never touches your machine; you're calling it over the wire through a reseller proxy and paying per clip.

The name is worth parsing, because it tells you a lot: "fal" means it goes through the fal.ai queue API, not some private endpoint. The pack's ComflyFalBase submits your job to fal-ai/pixverse/v6/image-to-video, then polls until the result is done and pulls the finished video back. Under the hood it's talking to https://ai.t8star.org/fal - the zhenzhen shop's proxy that stands in for queue.fal.run - and your api_key is the token you get from that shop, not a fal.ai key.

The inputs that actually matter

• prompt - required. Describe the motion, not the frame: "A cinematic camera move with subtle motion" is the default and it's a decent starting point.
• image (or image_url) - this node is image-to-video only, so you must give it a frame. In a real workflow you'd wire the output of a local text-to-image node straight into it.
• resolution - 360p up to 1080p, 720p default. Higher costs more and takes longer.
• duration - 1–15 seconds, 5 default.
• style - anime, 3d_animation, clay, comic, cyberpunk, or none. Fun, but I'd call it a novelty knob; the prompt does most of the work.
• generate_audio_switch and generate_multi_clip_switch - off by default; the first asks for synced sound, the second for multi-clip output.
• poll_interval / max_poll_attempts - how often it checks the queue and how long it waits. Defaults are 6s × 600 attempts, i.e. a one-hour timeout. Video queues back up; if your clip is still pending at the cap, raise the attempts before blaming the node.

Outputs are video (a real VIDEO tensor you can drop straight into ComfyUI's Save Video node - this pack went to the trouble of giving you an in-graph video, not just a URL), plus video_url and the raw response JSON if you ever need to debug or re-download.

Installing it

This is an API pack - no checkpoints, no VRAM required. Install with ComfyUI Manager (search "Comfyui-zhenzhen") or:

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

then restart ComfyUI. The requirements.txt pulls in requests, aiohttp, openai, torch, numpy, Pillow, opencv-python and friends - Manager handles most of it. You'll need an account + token from the zhenzhen shop (overseas: https://ai.t8star.org, domestic low-price: https://api.seedance.nz), pasted into the node's api_key field. FAL routes here are prepaid - credits are deducted up front and the difference refunded after the job finishes, and the final cost depends on resolution, duration and input size.

Common issues

• 443 / connection errors - the overseas shop is hosted outside mainland China; from a CN network you need a VPN with TUN mode. The pack has auto-failover between .org and .cn endpoints since mid-2026, which covers some of this.
• Timeouts - PixVerse queues fill up. Check max_poll_attempts before you assume it's broken.
• Seed cap - FAL seeds max out at 65535, and 0 means random. That's a platform limit, not a bug.

It's not the node for everyone - if you want local video, Wan and LTX exist and the external API node essay in our KB lays out exactly why people go either way. But if the model you want is PixVerse V6 and your machine can't run it, this is the least-friction way to get it inside a ComfyUI graph.