工具｜提示词选择器·图片

If you've found the full Prompts Selector and thought "I only generate images, why am I scrolling past video cases" - this is the version for you. The Image Selector is the same picker node pre-filtered to image-generation cards only. The brief shows ~336 choices, mostly avatars and virtual-identity portraits at the top of the list, but the filter is driven by card tags, so the whole image set (product shots, posters, style transfers) is in there. Same mechanism, same outputs, one less category of noise.

It's a pure UI convenience, not a generator: no API, no key, no network. It reads the locally-synced prompt library and serves cards. You pick, it hands you a prompt string, you wire that into a Tikpan image node like Qwen-Image-2.0 or Wan 2.7 Image Pro.

How it works

Under the hood it's the same class as the all-inclusive selector, with CARD_TYPE_FILTER = "image" - so the card list is scoped to cards tagged image before the dropdown is built. That scoping also narrows the 过滤_仓库 options to only the repos that actually contain image cards (the brief shows Nano Banana and GPT-Image-2 prompt collections) and drops the redundant image tag from 过滤_标签. 搜索关键词 and 显示详情 behave identically to the full selector.

What you actually notice day-to-day: the dropdown labels read like a catalog - "头像/虚拟形象 - 电影感柔和粉彩棚拍肖像", "电商商品图…", each with a short prompt preview so you can scan fast. The cards carry the full template text, including any {argument} placeholders you need to fill.

Inputs and outputs that matter

选择提示词 is the dropdown. Three STRING outputs: 提示词文本 (the card's prompt - wire into an image generator's prompt input), 卡片标题 (what the card is called), and 卡片详情 (full card body). Filter dropdowns 过滤_仓库/过滤_标签 and 搜索关键词 exist to shrink the list; 显示详情 toggles detail output.

Installing

Pack install as usual - Manager search "Tikpan", or git clone https://github.com/htrert/ComfyUI-Tikpan-Pro into custom_nodes, restart. Prerequisite, same as every library node: sync the library first with the Prompts Manager's 同步全部仓库 action, or the dropdown stays empty.

Common issues

• Empty dropdown. Unsynced library. Run the manager once.
• A card references a model you don't have. Many of these prompts were written for Nano Banana or GPT-Image-2 specifically. The library repos are model-flavored, so a "nano-banana" tagged card may underperform in Grok or Qwen. Check the tag filter, and remember prompt quality is model-relative (prompt-engineering.md) - the same token sequence that sings in one model can fall flat in another.
• {subject} placeholders. Several cards expect you to substitute a subject before generation. Literal braces in a prompt will not help you.
• Identity-reference cards need an actual reference image. Some avatar cards say "use my uploaded face image as the identity reference" - those expect a reference-image input on the generator node, not a text-only prompt.