- 提示词文本
- 卡片标题
- 搜索结果列表
- 结果总数
The Tikpan prompt library holds hundreds of curated cards - avatars, product shots, cinematic video cases, seedance styles, AI-director scripts. Scrolling a dropdown through 500 entries is a fool's errand. The Prompts Search node is the lookup tool: you type a keyword, it filters the local library across repo and tag, and hands you the matching prompt, its title, and the full result list. The workflow difference is real - instead of hunting through a giant enum, you ask the library a question.
It's a pure offline utility. No API key, no cost, no data leaving your machine - the prompt cards are already on disk from the manager node's sync, and search just reads them.
How it works
The mechanism is a straightforward filter over the synced card index. 搜索关键词 matches against card titles, prompt text, and card body (case-insensitive). 过滤_仓库 narrows to one of the nine source repos, 过滤_标签 to a tag bucket (director, gpt-image-2, image, nano-banana, video, seedance, guide, orchestration, video…). 最多返回数量 (default 10, up to 50) caps the result list. 结果索引 (0–49) is the subtle one: search returns a list, and this index picks which item from that list becomes the primary output. In practice you set it to 0, look at what the search list returned, then bump it if the interesting hit is further down.
Inputs and outputs that matter
Beginner's flow: type a keyword in 搜索关键词, run, and read the 搜索结果列表 (STRING) to see what matched. Then set 结果索引 to the row you want and read 提示词文本 (STRING - the actual prompt, ready to wire into a generator's prompt input) and 卡片标题 (STRING, the card's name so you know what you grabbed). There's also a 结果总数 INT output if you're building something programmatic. The README's canonical wiring is [提示词选择器] → 提示词文本 → [GPT Image 2 生图]; the search node plugs into that same slot.
Installing
Standard pack install - Manager search "Tikpan", or git clone https://github.com/htrert/ComfyUI-Tikpan-Pro into custom_nodes, restart. Then remember the one prerequisite: run the Prompts Manager node with 同步全部仓库 first. Search over an empty library returns nothing and the UI shows the "(请先同步提示词库)" hint.
Common issues
- Search returns nothing. Either the library isn't synced yet (run the manager), or your keyword doesn't appear in any card's title/prompt/body. The cards are mostly Chinese-titled with English prompts after translation, so try both languages.
- Wrong result for the index. The dropdown and search index are separate paths; if the prompt text doesn't match what you expected, you're reading the wrong row - check 结果索引 against the list.
- The library is stale. Search is only as current as the last sync. If you know a repo has newer cards, re-sync before searching. This is a content-freshness issue, not a search bug.
- Good prompts still need a human eye. A curated card is a starting point, not a license to stop thinking - the library's own docs frame these as templates to adapt, and prompt-engineering wisdom still applies: the card that reads perfectly on paper can still fight a specific model's vocabulary (prompt-engineering.md).
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| 搜索关键词 | STRING | — | |
| 过滤_仓库opt | COMBO | 全部 | 10 options: 全部, YouMind-OpenLab/awesome-nano-banana-pro-prompts, EvoLinkAI/awesome-gpt-image-2-API-and-Prompts, EvoLinkAI/awesome-seedance-2.0-prompts, EvoLinkAI/awesome-seedance-2-guide, toki-plus/ai-mixed-cut, +4 |
| 过滤_标签opt | COMBO | 全部 | 9 options: 全部, director, gpt-image-2, guide, image, nano-banana, +3 |
| 最多返回数量opt | INT | 101–50 | — |
| 结果索引opt | INT | 00–49 | — |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| 提示词文本 | STRING | — |
| 卡片标题 | STRING | — |
| 搜索结果列表 | STRING | — |
| 结果总数 | INT | — |