图片｜Grok Imagine Image 生图

This node is xAI's grok-imagine-image model dropped into a ComfyUI canvas - text prompt in, image tensor out, nothing running on your machine. It's part of the ComfyUI-Tikpan-Pro pack from Tikpan (攀升AI), which is a reseller layer that wraps closed cloud models behind one sk- API key. No weights to download, no VRAM needed, and that's the whole pitch: if you want Grok Imagine images in your workflow and you don't have (or don't want to spend on) the hardware, this is the on-ramp.

Understand what you're getting first, because the tradeoffs are real. Every call costs money (Tikpan lists this node at about 0.208 RMB per image), your prompt leaves your machine and lands on someone else's server, and the content filter belongs to the model, not to you - nothing to abliterate on a closed service. The community's honest line, per the KB, is that an API node is the right tool for a model you can't run locally and the wrong default for one you can. Grok Imagine is one you can't run locally, so this qualifies.

How it works

Mechanically it's a thin HTTP client. You fill in 生成指令 (the prompt), the node POSTs to Tikpan's /v1/images/generations endpoint with n/aspect_ratio/resolution/response_format params, waits for the result, decodes it, and hands back a standard IMAGE tensor plus a Log string. The model name grok-imagine-image is fixed; the node doesn't let you swap models, which keeps things simple.

The inputs that actually matter, out of the defaults:

• API_密钥 - your Tikpan key, starts with sk-, from https://tikpan.com. There's no free tier; you're on prepaid virtual tokens.
• 生成指令 - be specific. The default is a cinematic product-poster prompt, which tells you the author's target audience (this pack is built for TikTok e-commerce sellers). English recommended.
• 画面比例 - auto lets the model decide; 1:1, 16:9, 9:16 are the ones you'll actually set.
• 清晰度 - 1k is the cheap-and-fast lane, 2k doubles the detail and the bill.
• 生成张数 - 1 to 10; each one is another billed image.

Two optional toggles worth knowing: 跳过错误 returns a black image instead of killing the workflow when the API fails (handy for batch runs), and 返回方式 switches between a cloud URL and raw Base64 - leave it on url, it's what the Image output expects anyway.

Wiring it up

Outputs are Image (IMAGE) and Log (STRING). Plug the Image into a Save Image / Preview node like you would any generator. That's the whole appeal: it sits in your graph next to your local upscaler and masks, letting the closed model do the part you can't.

Install and first run

Install is standard for the pack:

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

Restart ComfyUI, then either use ComfyUI Manager (search "Tikpan") or the clone above. No extra Python deps beyond what ComfyUI already ships - the code uses requests, numpy, torch, and Pillow. No model downloads at all. Then grab your key from tikpan.com and paste it in.

Common issues

The usual failure modes are key and money, not code. A 401 means the key is wrong or has stray whitespace; 402/insufficient_quota means you're out of balance; 429 means you're hammering it - back off and wait 30–60s. If the request dies mid-batch, the Log output will show you the raw error and whether you still got billed. Start at 1k and one image to test, then scale up once the prompt is right - that habit will save you more than any single setting.