new.ip233.com｜Grok 通用视频生成

This is the flexible Grok video node. Where its 1.5 sibling demands exactly one reference image, TikpanCangyuanGrokVideoNode does text-to-video, single-image, multi-image with up to seven reference frames, and even video editing if you feed it a 视频URL. Same relay, same /v1/video/generations endpoint, but a much broader brief - this is the one you reach for when you want Grok's motion with real character/subject consistency from a stack of stills, or when you want to restyle an existing clip.

It lives in ComfyUI-Tikpan-Pro (the official plugin for Tikpan.com), in the new.ip233.com relay group - so it needs its own sk- key from https://new.ip233.com. Closed model, pay-per-video, no local weights involved.

How it works

Same shape as the other relay video nodes: prompt + optional image(s) get POSTed to the relay as JSON, the node polls until the job finishes, downloads the MP4 to your output folder, and returns a VIDEO tensor plus paths and logs. The differences are in what you're allowed to send: up to 7 reference images (参考图1–7), a 视频URL that flips the call into video-edit mode, and a prompt cap of 4096 characters - enough for a detailed scene description.

The inputs that matter

• 生成指令 - the prompt. Keep it under 4096 characters; the relay hard-caps it.
• 视频时长 - 4 to 15 seconds in the same 4/6/8/10/12/15 dropdown.
• 分辨率 - 480p or 720p.
• 画面比例 - the widest set in this relay group: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3.
• 参考图1 … 参考图7 - optional image inputs. Zero for pure text-to-video, a few for subject consistency, seven if you're being ambitious. Keep the subject in frame in every still - that's what actually drives consistency, not the model's hope.
• 视频URL - optional; fill it to run in video-edit mode (a URL Grok's servers can fetch). The tooltip is explicit that this changes the call into video editing.

The usual relay tail - 最长等待秒数 (300), 查询间隔秒数 (5), 校验HTTPS证书 (off), 跳过错误 - is unchanged.

Outputs

Same five as the 1.5 node: 视频输出 (VIDEO), 本地保存路径 (MP4 in the output dir), 视频云端直链, 任务ID, 完整日志.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/htrert/ComfyUI-Tikpan-Pro

or ComfyUI Manager → search Tikpan → Install → restart. Stock deps only.

Gotchas

Two traps. First, multi-image mode is where people overload the model - throwing seven unrelated images at it makes the output muddier, not more consistent; two or three well-composed frames usually beat a pile of random stills. Second, the video-edit mode needs a publicly reachable URL; a localhost or intranet link won't work because the relay's servers are the ones fetching it. And the standing reminder for this whole pack: it's a thin-documented relay plugin with basically no community discussion, so test on short 4-second 480p clips before you spend on the long 720p renders.