SuperAPI AI Directory
Partner中文AI newsFavorites
Skillstweet-style-importer

tweet-style-importer

Design & creation

View repository

Updated 10-16 15:13

About this skill

Tweet Style Importer (Tweet Style Auto-Ingest Skill)

This Skill is used to automatically fetch, process, and standardize AI image generation style tweets on X (Twitter) (such as Midjourney --sref styles, artist portfolio image sets, etc.) into the hand-drawn style library with one click.

Core Automated Workflow

When the user provides a tweet link (for example, https://x.com/.../status/...) and requests ingestion, execute the following 6-step standard process:

[1. 抓取推文与高清原图]
       ↓
[2. 计算下一位顺延编号 (如 #217)]
       ↓
[3. 生成双轨图像资源]
   ├── 生图参考图：多图时为 1024×1024 四宫格；仅一张原图时直接使用单图
   ├── 风格单图：512×512 纯净风格代表图，严禁压印任何数字编号、作者Handle或角标框 (201-400/217.webp)
   └── 画廊展板：1254×1254 16 宫格画板（#201–216 使用 G；#217 起使用 H，单图以纯净画面进入展板，不压印角标）
       ↓
[4. 风格库元数据入库 (styles_200_reorganized.md)]
       ↓
[5. 模型能力策略配置 (model_capabilities.json)]
   └── 强制启用传图规则：name_activation=none, traits_activation=none
       ↓
[6. 重建画廊并运行全量校验 (build_library.py & validate_library.py)]

One-Click Execution Command

Run the automation script directly in the workspace to complete the entire process:

# 标准一键执行
python scripts/import_tweet_style.py "https://x.com/username/status/..."

# 支持可选的自定义英文风格名与中文特征描述
python scripts/import_tweet_style.py "https://x.com/username/status/..." --name "Custom Style Name" --traits "核心特征描述"

Specifications and Output Standards for Each Step

1. Fetch Tweet and High-Resolution Original Images

  • Before making a network request, first check by X/Twitter status ID downloads/tweet_{id}/tweet.json; for tweets already fetched, stop directly, do not download images, do not assign new numbers, and do not modify the style library or gallery.
  • Automatically parse the tweet author (name and Handle), tweet body, and Midjourney --sref style code.
  • Automatically extract all original images (force replacement with ?name=orig high-resolution size).
  • Data is saved in downloads/tweet_{id}/(including tweet.json and the original images).
  • After successful ingestion, write to import.json, recording the status ID, original link, style number, and import time; on duplicate input, return the already-ingested number. Without import.json , historical downloads only prompt "already fetched but registration not completed" and likewise will not be fetched repeatedly.

2. Image Specification Standards

  • Image generation reference images：
    • When there are 2–4 different original images, create a images/individual/{bucket}/{number}_grid.webp：1024 × 1024 px JPG 2×2 four-panel grid.
    • When there is only 1 original image, do not create _grid.webp, directly use images/individual/{bucket}/{number}.webp, and do not copy it into four identical panels.
  • Style single image (images/individual/{bucket}/{number}.webp)：
    • Specifications:512 × 512 px WebP
    • Details: clean image,strictly forbidden to imprint any digital number, author Handle, or corner label frame, fully preserving the original artwork content and composition details.
  • {bucket} increments every 200 numbers:001-200、201-400、401-600. When a four-panel grid exists, it is located in the same directory as the single image with the same number.
  • Gallery display board (from #217 onward, images/H_{number}.webp）：
    • Specifications:1254 × 1254 px WebP
    • Layout: standard 4×4 (16-panel) board, filled left to right, top to bottom; do not create a new board until full.
    • Naming: the filename reflects the range of numbers actually filled in, for example H_217-218.webp; only after 16 panels are full does the next number create a new board.
    • Status:references/contact_sheet_state.json records the active board, number filled, and next panel; a single image enters the corresponding panel as a clean image without corner labels.

3. Style Library and Model Strategy Ingestion

  • Main style table (styles_200_reorganized.md)：
    • Update the total count title and append a new row: | {number} · {author} | {generation_name} | {traits} |
  • Model capability strategy (references/model_capabilities.json)：
    • Append a forced image upload configuration for this number: "{number}": { "name_activation": "none", "traits_activation": "none" }
    • Ensure that when calling resolve_reference.py , regardless of model, it is always determined as use_reference_image = True, and prioritizes returning the multi-image four-panel grid; for single-image imports, return the single-image path.

4. Gallery and Verification

  • Call python handdraw-style-prompter/scripts/build_library.py to rebuild the gallery and JSON index.
  • Call python handdraw-style-prompter/scripts/validate_library.py to verify the full-library contract, ensuring no number gaps and no missing images.

Delivery and Reporting Format

After execution is complete, output the following standardized result card:

  1. Ingestion information: number, author, style name, and group it belongs to.
  2. Gallery link：skills/handdraw-style-prompter/gallery/index.html。
  3. Bilingual prompt template：
    • Chinese:风格名称：#{编号} · {generation_name}。主题：[主题]。参考作者/风格名称：{author}。
    • English:Style name: #{number} · {generation_name}. Theme: [Theme]. Reference author/style name: {author}.