图片｜Qwen-Image-2.0 生图/编辑

Qwen-Image is Alibaba's image-model family - and by the time this pack shipped, community interest in it was huge (the "qwen image" phrase saturates the Reddit corpus at 1000+ threads). The Qwen-Image-2.0 node puts that model inside your ComfyUI graph without you downloading a single weight: it's an API wrapper over Tikpan's /v1/images/generations endpoint. Type a prompt, get an image tensor back, wire it into a Save Image node. From the canvas it looks like any other generator; underneath it's an HTTP call with your key baked in.

That's the whole trade, and it's worth being honest about it up front: this node has no local VRAM cost and installs in seconds, but every image is metered, and your prompt (and any reference images) leave the machine for Tikpan's servers. For a model you can't run locally - or won't wait to download - that's the deal (external-api-nodes.md).

How it works

The node posts a standard OpenAI-style image-generation request to POST /v1/images/generations with the model ID qwen-image-2.0-2026-03-03, then decodes the returned image into a ComfyUI IMAGE tensor. The visible fields map almost one-to-one onto that request: 生成张数 is the n parameter, 水印 is watermark, 提示词扩展 is prompt_extend. The two dropdowns that look like choices - 画面比例 (auto/1:1/16:9/9:16/…) and 清晰度 (auto/1k/2k) - aren't sent raw; the node uses them to compute a pixel size (like 1152x2048), a pattern the pack uses across its image nodes. 模型 is fixed to the single Qwen-Image-2.0 variant this node wraps.

Inputs and outputs that matter

For a beginner: API_密钥 (your sk- key from tikpan.com), 生成指令 (the prompt - tooltip suggests English, and specificity really does matter with Qwen-Image's text-rendering), and 画面比例 + 清晰度 for the frame you want. 生成张数 lets you get up to 4 images per call. The optional 跳过错误 is the batch-flow friend: when on, a failed request returns a black image instead of killing the whole workflow.

Outputs are Image (IMAGE - the result) and Log (STRING - request summary and any error text). Image → Save Image; Log → a text viewer if you're debugging.

Installing

Standard pack install - ComfyUI Manager search "Tikpan", or:

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

Restart, grab a key at tikpan.com, paste it in. No model download, no VRAM requirement - the pack's own banner pitches it as "0.6元≈1美金余额" prepaid tokens.

Common issues

• Black image + "API key is empty." The node validates the key length (sk- alone is rejected). Pasted-but-invalid keys produce a similar failure - check the Log output.
• "2K" renders that come back smaller. This was a real bug in this pack's history: passing a string like "2K" instead of pixels made some models silently return default resolution. The fix (pixel conversion + Lanczos fallback) is in the changelog - if you still see it, it's an upstream/API quirk, and the Log will show what size was actually requested.
• Cost creep on batch runs. n up to 4 multiplies the bill per call. The tooltip on 生成张数 is a reminder for a reason - check it before a big batch.
• Security posture. It's an API node holding your key and calling the network by design. Same frame as every wrapper node: prefer vetted installs, read before trusting, be most suspicious of "one node, paste key, it just works" (external-api-nodes.md § Security).