图片｜Wan 2.7 Image Pro 生图/编辑

Wan is Alibaba's generation-model line - the name most people know from its video models, but the image branch is very much alive. The Wan 2.7 Image Pro node is Tikpan's wrapper around it, with the pack's marketing calling out 4K resolution and deep "thinking" reasoning modes. You get the model in your graph through Tikpan's /v1/images/generations endpoint - no weights to download, no VRAM check, just a key and a prompt. It's the sibling of the Qwen-Image-2.0 node in this pack, and the honest framing is identical: an API wrapper is the only way to use a model you aren't going to download, and every render is metered and leaves your machine (external-api-nodes.md).

Where this node stands out from its sibling: the 清晰度 dropdown goes to 4k, and the default resolution is 2k rather than auto. That's the headline feature - if you need big, detailed, text-heavy output for print or high-res e-commerce listing pages, this is the Tikpan image node with the headroom.

How it works

Same mechanics as the pack's other image node: POST to /v1/images/generations with model wan2.7-image-pro, decode the response into a ComfyUI IMAGE tensor. 生成张数 maps to n, 水印 to watermark, 提示词扩展 to prompt_extend. 画面比例 (auto/1:1/16:9/…) and 清晰度 (auto/1k/2k/4k) combine to compute a pixel size for the request - auto falls back to the model default. The default prompt is already tuned for the "high-end commercial product shot with 4K detail" use case the pack targets, which tells you who this is built for: e-commerce content pipelines.

Inputs and outputs that matter

Set API_密钥, write the 生成指令, pick 画面比例 and 清晰度. The two knobs worth actually thinking about: 清晰度 = 4k costs more and renders slower - 2k is the sane default, reach for 4k only when the output genuinely needs it - and 生成张数 up to 4 multiplies the bill per call. 跳过错误 (optional) swaps failures for black images so a batch doesn't die mid-run.

Outputs: Image (IMAGE) and Log (STRING). Image → Save Image / Preview; Log → text viewer for debugging.

Installing

Pack install as usual - Manager search "Tikpan", or:

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

Restart, get a key from tikpan.com (prepaid virtual tokens, "0.6元≈1美金余额" per the node's own banner). Nothing to download locally.

Common issues

• Black image + error in Log. Key problems first: blank/placeholder key gets rejected by validation. After that, check whether 4k is supported by your chosen aspect ratio - a 4k × ultra-wide combination can exceed the endpoint's limits. Drop to 2k and retry.
• "4K" that isn't 4K. Same historical bug family as the pack's other image nodes - resolution strings were once passed through instead of pixel values, silently yielding default sizes. The changelog documents the pixel-conversion fix; if output still looks low-res, read the Log's requested-size line.
• Thinking mode costs. The pack pitches "deep thinking reasoning" as a feature; if the node or API exposes it, expect slower and pricier renders than the plain path. For straightforward product shots you usually don't need it.
• Security posture. API wrapper holding a key and phoning home by design - the standard caution applies: read a fresh install, don't reuse keys across untrusted nodes (external-api-nodes.md § Security).