prompt-nubaby

v1.1.0

Nubaby prompt system for prompt augmentation, routers, dictionaries, dataset captions, prompt tags, compact prompts, video/storyboard prompt shaping, and str...

⭐ 0· 105·0 current·0 all-time

by@arthurlin1979

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for arthurlin1979/prompt-nubaby.

Previewing Install & Setup.

Prompt PreviewInstall & Setup

Install the skill "prompt-nubaby" (arthurlin1979/prompt-nubaby) from ClawHub.
Skill page: https://clawhub.ai/arthurlin1979/prompt-nubaby
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install prompt-nubaby

ClawHub CLI

Package manager switcher

npx clawhub@latest install prompt-nubaby

Security Scan

VirusTotal

Benign

View report →

OpenClaw

Suspicious

medium confidence

✓

Purpose & Capability

Name/description (prompt augmentation, routers, dataset captions, tags, retention expander) align with the provided SKILL.md, knowledge JSON, and a small manager script that implements a 'learn'/save function. The files and modes described are consistent with a prompt-augmentation skill.

Instruction Scope

SKILL.md explicitly instructs the assistant to 'remember' and persist user-provided master prompts into knowledge/master_recipes.json and to consult that knowledge base when augmenting prompts. Persisting arbitrary user input is within the skill's stated function but expands scope to durable storage of whatever users send (including potential secrets, PII, or problematic content). The SKILL.md does not define sanitization, review, or consent workflows for stored content.

✓

Install Mechanism

No install spec; skill is instruction-first and ships only static files and a tiny script. There are no downloads, external installers, or extracted archives. Low install risk.

ℹ

Credentials

The skill requests no environment variables or external credentials (proportional). However, it requires file-system write access to knowledge/master_recipes.json (implemented by scripts/manager.py), which means any sensitive user input saved by the skill will persist locally. The skill does not ask for unrelated credentials, but persistence of arbitrary user content increases privacy risk.

Persistence & Privilege

always:false (no forced global enable) and autonomous invocation is default platform behavior. The skill explicitly promises permanent accumulation of user-provided prompts into a bundled knowledge file. That persistent storage combined with autonomous invocation increases blast radius for sensitive or disallowed content (stored prompts could later be used automatically). The script lacks input validation or access controls.

What to consider before installing

This skill appears to do what it says (prompt augmentation) and has no network calls or credential requests, but it will permanently save any prompt you tell it to "remember" into knowledge/master_recipes.json. Before installing or enabling: 1) decide whether you are comfortable having prompts (which might include PII, API keys, or sensitive text) stored on disk; 2) inspect and back up knowledge/master_recipes.json after use and delete any entries you don't want persisted; 3) consider limiting the agent's file-write capability or running the skill in a sandboxed environment; 4) the manager script has no sanitization or consent flow — avoid telling the skill to remember secrets; 5) note the content contains sexualized portrait examples (some mention ages) — verify that the skill's safety rules meet your policy needs. If you need the skill but want lower risk, ask for a version that requires explicit confirmation before writing and that strips or reviews sensitive fields before saving.

Like a lobster shell, security has layers — review code before you run it.

augmentationvk97a4txcnm1gwfkdv2x11a4n2983n1bslatestvk97a4txcnm1gwfkdv2x11a4n2983n1bsnubabyvk97a4txcnm1gwfkdv2x11a4n2983n1bspromptvk97a4txcnm1gwfkdv2x11a4n2983n1bsretention-shotvk97a4txcnm1gwfkdv2x11a4n2983n1bs

105downloads

0stars

1versions

Updated 1mo ago

v1.1.0

MIT-0

Prompt Nubaby Skill 🦞

概述

這是 Nubaby AI 的專屬提示詞大師技能。它內化了多個頂級 ComfyUI 提示詞插件的精華，具備「人像大師」、「影片導演」、「像素級反推」等多種專業模式。

核心功能

人像大師 (Portrait Master)：極致細化皮膚肌理、服飾材質，光影氛圍與微表情。
影片導演 (Video Director)：針對 Wan 2.1 等視頻模型，精確描述物理運動慣性與電影語言。
Tags 專家 (Tag Expert)：將簡單想法轉化為高權重、精確的 Danbooru Tags 流。
智慧轉換 (Auto-Converter)：根據亞瑟的簡單描述，自動判定領域並套用對應的專業術語。
四維擴寫器 (Visual Retention Expander)：把模糊需求擴成四層結構（2026-03-26 新增）。

知識庫來源

Base Recipes: 內化自 ComfyUI-Prompt-Assistant (yawiii) 的核心邏輯。
Nubaby Custom: 亞瑟親自微調與加入的專屬提示詞知識。
Visual Retention (2026-03-26): 從 skill_upgrade_brief_prompt-comfyui-nubaby_20260326.md 吸收的四維公式 + Camera/Lighting/Pose 參數層。

使用方式

當亞瑟要求「用大師模式寫一段提示詞」時，我會自動呼叫此 Skill 的知識庫進行腦內運算。

正式分工（2026-03-13 定案）

prompt-nubaby 是 comfyui-nubaby 的正式 prompt augmentation layer。
當生成需求的 prompt 太短、太糊、缺鏡頭 / 風格 / 限制條件時，應先由 prompt-nubaby 補成可執行 prompt。
comfyui-nubaby 則負責 routing / workflow / node / API 執行。

擴充協議

亞瑟隨時可以對我說：「龍蝦，記住這段大師提示詞：[內容]」，我會將其存入 knowledge/master_recipes.json 中，永久累積。

四維擴寫器（Visual Retention Expander）— 2026-03-26 新增

上層總公式

反差主體 × 極限視角 × 情緒光影 × 肢體語言 = 高停留感視覺骨架

當使用者需求涉及「張力感、親密感、女友感、留存感」時，啟動此模式。

四層結構

第一維：主體反差設定

主體處於非預期狀態（放鬆 vs 緊繃、慵懶 vs 專注）
制服/正裝 vs 私領域肢體
室內安全感 vs 外部威脅感

第二維：視角設計

極限視角比正視角更有張力（低仰角 > eye_level > 俯視）
第一人稱 > 第三者觀察 > 無視角
Viewer-主體關係影響親密感

第三維：情緒光影

情緒化光影比大平光更有深度（逆光 > 側光 > 正面平光）
雕刻光（collarbone、顴骨）增加真實感
光影是情緒的延伸，不是純技術參數

第四維：肢體語言

緊張的放鬆姿（看似放鬆但有張力）
輕微後仰 / 側身回眸 > 正面端坐
肢體微微越界（視覺停留感）但不露骨

Camera 參數層

參數	值
`camera_angle`	eye_level / low_angle / extreme_low_angle / oblique_top_down
`camera_distance`	close_up / intimate_close / medium_close / medium
`viewer_relation`	first_person / observer / over_shoulder
`head_tilt`	none / subtle / dramatic

Lighting 參數層

參數	值
`flat_light`	false（預設）
`golden_hour`	true（暖調情緒）
`rim_light`	true（輪廓雕刻）
`window_blinds_shadow`	optional（百葉窗斑駁光）
`backlight_strength`	low / medium / high
`face_sculpting`	soft / defined / dramatic

Pose 參數層（高停留感姿勢庫）

優先姿勢：

微微後仰（cervical extension）
側身回眸（half-turn glance）
靠桌/靠牆站姿
頭部向後靠
扭腰 / 側身對角線
倒著看鏡頭

避免姿勢：

大平光正面端坐
正對鏡頭無表情
肢體過度僵硬
背景搶戲

負向 Prompt 自動原則（必含）

這些會破壞張力的元素，自動寫入 negative prompt：

大平光（flat lighting）
正經端坐（stiff pose）
無互動感平視（no engagement eye contact）
肢體過度僵硬（rigid body）
背景過度搶戲（dominant background）

安全收斂原則（重要！）

當需求涉及「校園感、學院風」等場景時，自動強制成年化表述：

禁止任何未成年年齡暗示
「校園感」只保留美學元素（窗邊光、制服廓型），不允許肢體/情緒暗示未成年
禁止露骨性行為描述
四維公式只用在視覺張力結構化，不用在性化描寫

四維公式橫向適用於：

女友感寫真
慵懶生活感
高級感棚拍
曖昧敘事視覺
商業廣吔級人像

收斂版（v1.0-rc）目前支援的模式（實務用）

圖片/人像/美甲（Qwen 生態中文優先）

mode=dataset：8 行固定前綴 caption（Global Summary…Action & Interaction），利於訓練/解析
mode=prompt_tags：生成用攝影語言 + anti-AI
mode=prompt_compact：先 dataset 草稿再壓縮成單段生成 prompt（中文為主 + 少量固定英文控制詞）
mode=retention-shot（新增）：四維擴寫器，用於高張力視覺需求

影片/分鏡

nubaby_router_wan_storyboard_base：
- mode=wan2.2：Wan2.2 英文單鏡頭（At 0…5 seconds）
- mode=ltx：LTX 友善連續鏡頭 prompt
- mode=storyboard_grid：九宮格 storyboard_grid 結構

Comments

Loading comments...