福利｜Grok Video 1.5 单图生视频

This is the "benefit" (福利) channel for Grok Video 1.5 - a deliberately dumbed-down node that does exactly one thing: turn a single image into a video with grok-video-1.5. The author's framing is "welfare/benefit," i.e. a promotional lane, and the simplicity is the point. There's no model dropdown, no mode selector, no multi-reference ports. Connect one image, write a prompt, get a clip. If you want Grok Video 1.5 without tripping over the reference-count rules of the main Grok Video node, this is the one that can't be misconfigured.

The pack's CHANGELOG shows why it exists: the main 1.5 path had a real bug where the seconds parameter went out as a number and the relay rejected it with an invalid_json unmarshal error. This node was built (and then fixed in v1.4.5) as the clean, fixed-config route - worth knowing if you hit that error on the other node, because the answer is "use this one" or update the pack.

How it works

Same metered cloud API under the hood as every Tikpan video node: submit to the fixed /v1/video/generations endpoint, poll until the clip renders, download, hand back the five standard outputs - 🎬_视频输出 (VIDEO), 本地保存路径, 视频云端直链, 任务ID, and 完整日志. The one structural difference is the 任务ID output being present here, which is your recovery handle if a long job times out.

The inputs that matter

• 参考图 - required, and exactly one. The tooltip doesn't mince words: "必填且只能 1 张参考图."
• 生成指令 - how the still should move: "cinematic video, smooth natural motion, realistic lighting, coherent camera movement" is the shipped default and it's a fine starting point.
• 视频时长 - 4/6/8/10/12/15 seconds.
• 分辨率 - 480p or 720p.
• 画面比例 - 16:9 or 9:16 (no 1:1 here).
• 随机种子 - fixed at 888888 by default.

Install and first run

Pack-standard, and light - nothing to download, the model is in the cloud:

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

Restart (or ComfyUI Manager → "Tikpan"), paste your sk- key from https://tikpan.com into API_密钥, load one image, run. Because it's synchronous, the node blocks while polling - a 6s 480p clip is the right first test before you commit to 15 seconds at 720p.

Common issues

The whole point of this node is to have fewer failure modes, and it mostly succeeds. Watch for the standard trio - 401 bad key, 402 empty balance, 429 calling too fast - and for timeouts on long or 720p jobs: if it times out, hold onto the 任务ID and collect the finished clip with the pack's Task Fetcher instead of resubmitting and paying twice. And feed it a good image; with a single-image model, the input frame is the whole ballgame.