工具｜提示词库管理器

The Tikpan Pro pack ships a prompt library - a curated pile of prompt templates synced from nine GitHub repos covering Nano Banana, GPT-Image-2, Seedance, AI mixed-cut, highlight clips, video pipelines and more. The selectors and search nodes are the shopfront. The Prompts Manager is the loading dock: it's the node that actually pulls those repos down and builds the local card database everything else reads. If you add a Tikpan prompt selector and the dropdown only says "请先同步提示词库" (please sync the prompt library first), this is the node you were missing.

The good news: it's dead simple and free. No API key required for the basic sync. The optional key only buys you one feature - machine translation of the English prompts into Chinese as they're cached.

How it works

The manager reads a fixed list of nine repos (YouMind-OpenLab's nano-banana-prompts, EvoLinkAI's gpt-image-2 and seedance collections, toki-plus's video repos, shuyu-labs' BigBanana-AI-Director, etc.) and syncs each into ComfyUI/custom_nodes/ComfyUI-Tikpan-Pro/data/prompts-library/ as a JSON card index. There's no cleverness - it's a git-fetch-and-parse pipeline that turns repo content into labeled cards (title, prompt, tags, repo source) that the selector nodes filter on.

Two operating modes via the 操作 dropdown:

• 查看状态 (default) - prints a report: total cards, per-repo card counts, last-sync time, any sync errors. This is your "did it work?" screen.
• 同步全部仓库 - pulls everything (needs internet, ~10–30 seconds per the README).

同步单个仓库 narrows a full sync to one repo. The optional Tikpan_API密钥 + 翻译模型 (deepseek-v4-flash) pair translates English titles and prompt previews to Chinese during sync - nice if your team reads Chinese, pointless if you don't, and it does cost tokens so leave the key blank to skip it.

Output: a single 状态报告 STRING, human-readable, typically run once and forgotten.

Inputs and outputs that matter

For a beginner: set 操作 to 同步全部仓库, run it once, wait, then switch back to 查看状态 to confirm counts went up. That's the whole loop. The 翻译模型 enum only has one choice (deepseek-v4-flash), so don't overthink it.

Installing

Pack install as always - Manager search "Tikpan", or git clone https://github.com/htrert/ComfyUI-Tikpan-Pro into custom_nodes, restart. No heavy deps for this node specifically; it's stdlib JSON + HTTP.

Common issues

• Selectors are empty. You synced nothing yet. Run 同步全部仓库 once - this is the #1 "it doesn't work" moment with the prompt library, and the fix is running the manager first.
• Sync fails mid-way. The status report tells you which repo errored. GitHub repos get renamed or go private all the time; the 9-repo list is hardcoded in this pack version, so a dead repo is a pack-update problem, not something you can fix in-node.
• Rate-limit on sync. Pulling nine repos at once against GitHub's API can trip limits in quick succession. The manager handles per-repo errors gracefully; just re-run.
• Translation burn. If you paste an API key in, every sync with new content calls the translation model. It's cheap (deepseek-v4-flash is priced as a budget model), but it's also a network call with your key - same security posture as every other API node in this pack (external-api-nodes.md § Security). Leave the key empty unless you actually want the Chinese translation.