图片｜GPT-Image-2 福利生图

Every reseller has a loss-leader, and this is Tikpan's: the "福利" (benefit/welfare) node for GPT-Image-2, positioned as 源头拿货价 (source-at-cost pricing) in the header hint. TikpanGptImage2BenefitNode is a GPT-Image-2 text-to-image node that routes through a discount 福利 channel instead of the main one - same gpt-image-2 model, same /v1/images/generations call, cheaper meter. If you're doing high-volume, low-stakes generation - batch product backgrounds, moodboard drafts, social assets you'll never print - this is the node that keeps per-image spend down. It's the budget lane; the "official" node is the same model at standard pricing.

How it works

Synchronous call: build the payload (prompt, resolution tier, aspect ratio, quality, output format), POST to the benefit channel's /v1/images/generations, decode the image, save it, and emit 🖼️_生成结果图 (IMAGE) plus 📄_渲染日志 (STRING). Mid-graph output - wire it into a Save Image / preview node.

The inputs that actually matter

• API_密钥 - sk- key. The twist: the 福利渠道 dropdown picks which relay channel you use, and the hint says to use the API key corresponding to the selected channel. Channel one and channel two are two different meters - read the hint before pasting.
• 生成指令 - the prompt (default is an aurora/northern-lights scene, 8k cinematic).
• 模型 - gpt-image-2, only option.
• 分辨率档位 - 512 / 1K (1024, default) / 2K (2048) / 4K (3840). Higher = slower and pricier; 1K is the sensible bulk default.
• 画面比例 - nine options from 1:1 to 9:21.
• 画质与推理强度 - auto / low / medium (default) / high. This maps to the quality parameter.

Optional: 返回格式 (png / jpg / webp) and 跳过错误 - here it means network errors, insufficient balance, or API errors return a black image instead of killing the workflow, which is handy for unattended batch runs.

Installing it

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

restart (or Manager → "Tikpan"). No models to download - relay + decode/save.

Where people get burned

• It's a reseller discount channel. "福利" pricing is subsidized and can move. If your workflow depends on the rate, expect the pricing to change and have a fallback path (the official node is the same model at standard rates).
• Channel/key mismatch. Two 福利渠道, and each expects its own key. Wrong pair → auth error. The header hint is the map; follow it.
• Black images are the failure mode. With 跳过错误 on, a failed run returns a black image silently. That's the point (keep the pipeline alive), but it means you should spot-check output for black frames in batch runs.
• Same quality caveats as the official node. It's still GPT-Image-2 - the discount changes price, not capability or moderation.

The honest read: a legitimate way to burn through cheap GPT-Image-2 generations for bulk work. It's also the node that most loudly says "you're on a third-party relay," so keep the security lens from the KB's API-nodes essay on: the key lives in the node, the calls leave your machine, and the channel's discount is vendor-controlled.