图片｜GPT-Image-2-C 旧版多参考生图

Like its edit sibling, TikpanGptImage2GenNode is a compatibility node that keeps the old gpt-image-2-all multi-reference generation entry alive under the renamed model gpt-image-2-c. Where the official node is prompt-to-image and the V2 editor is edit-based, this one is the generation path with reference images: give it a brief plus up to five reference images, and the model creates new work that draws on them. For anyone whose saved workflow says gpt-image-2-all, this is the drop-in that makes it load again - and as a generation node it's also the more flexible of the two legacy GPT-Image nodes.

How it works

The prompt and up to five 参考图 go to Tikpan's image generation endpoint as a multi-reference request; the model creates a fresh image informed by the references (identity, style, composition hints), which comes back as 生成结果图 (IMAGE) plus 渲染日志 (STRING). There's a 随机种子 field (default 888888) for reproducibility - though, as with most reseller relays, treat it as a hint rather than a guarantee.

The inputs that actually matter

• API_密钥 - sk- from tikpan.com.
• 生成指令 - the brief (default: "Create a high quality image. Use any connected reference images as visual references."). Say what the references are for - product, style, character - so the model fuses them deliberately.
• 分辨率档位 - 512 / 1K (default) / 2K / 4K. Higher = slower, pricier.
• 画面比例 - 1:1 / 16:9 / 9:16 / 21:9 / 4:3 / 3:4.
• 品质 - standard (faster) or hd (default, better detail).
• 参考图_1 … 参考图_5 - up to five visual references.

Optional: 中转站地址 (defaults to tikpan.com, leave it).

Installing it

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

restart (or Manager → "Tikpan"). No models to download.

Where people get burned

• Five references is the cap here. The pack's Gemini node goes to 14; this legacy generation node stops at 5. Don't over-wire.
• Seed is a hint, not a promise. Same caveat as the sibling nodes - reseller relays don't always honor deterministic seeds. The field exists for API parity; don't build a reproducible-batch pipeline on it.
• Legacy lane, again. New work has the official node (clean prompt-to-image) and the V2 editor (multipart edit with 15 references + mask). This node's real job is keeping gpt-image-2-all graphs alive. If you're starting fresh, weigh whether multi-reference generation on the new endpoint is a better base than a renamed legacy model.
• References are data too. Every reference image leaves your machine for the relay. Same privacy calculus as everything in this pack.

The fair summary: a solid, simple multi-reference generator that happens to be a compatibility relic. If your old workflow references gpt-image-2-all, this is the node you want - and the pack's own structure is telling you where the new investment went.