R0xyHaruhi/CreatorLab

Overview

CreatorLab is a local workbench for individual Chinese content creators, focused on Douyin and Bilibili. It imports a creator's own post data, visualizes performance, and supports a review loop: inspect a post's real metrics, choose one change, then compare later posts against it. It is aimed at solo creators rather than teams or MCN organizations.

The workbench has four entry points: viewing works (battle-power charts, conclusions, reports and PNG export), planning the next post (evidence-based suggestions, local video checks, experiments and reviews), a searchable work library, and sync/settings. Bilibili data comes from public submissions read by UID; Douyin data comes from the creator center and requires login plus ownership confirmation.

Getting started (Windows)

1. Install 64-bit Python 3.12 with the Python Launcher, unzip the source, and double-click start_windows.bat. The first run installs dependencies and needs network access.
2. Use the built-in sample without logging in, or open “Sync & Settings” to add an account. For Douyin, a nickname alone can be entered, but post ownership must be confirmed after syncing; for Bilibili, enter a UID.
3. Read one work in “View works”, then move to “Next post” to select one change, record it, and review it against later works.

Keep the launch window open while using the app; press Ctrl+C to exit. repair_windows.bat repairs dependencies and recover_windows.bat restores data. Detailed first-run steps and common blockers are in First run.

For development, the README documents a Python 3.12 virtual environment, pip install -r requirements-dev.txt, pytest -q, and python -m streamlit run app.py; the Streamlit server listens on localhost only and is not intended as a public cloud service.

Constraints and data handling

• Windows has received local-environment installation and first-use validation; macOS and Docker have no machine-tested acceptance in this round, and Docker only covers analysis and file import, not browser collection.
• Douyin import depends on the creator center and may be affected by page changes, login, or verification. Optional scheduled updates apply only to confirmed, uniquely owned works and require the service to stay running; they pause when login or verification fails. Bilibili has no equivalent scheduled-update entry.
• Bilibili public videos generally lack completion rate, and the tool does not estimate it. Missing values are not filled with zero, and metrics use their own valid samples and observation windows.
• Personal data, including the database, covers, backups, and browser login sessions, is stored under local/ by default. Do not upload or share the whole working directory; use the publishing flow to produce a clean source package.
• Cloud AI models are off by default and require the user to configure DeepSeek or Doubao; requests are sent only after preview and explicit consent and may incur provider fees (Chinese AI setup).
• The project is MIT-licensed, with fonts, icons, and project skills retaining their own licenses.

HotRepo's take

This is a small, early project: 127 stars and 5 forks as of the 2026-11-15 snapshot, with no recorded GitHub daily ranking. In the new-project Python list on that date it gained 27 stars over a roughly 12-hour window, growing from 100 to 127 stars. Those numbers describe initial attention, not adoption or data quality. What stands out in the materials is the emphasis on measurement discipline—separate metric samples, no zero-filling, no estimating Bilibili completion rates, and an explicit statement that correlation is not causation, along with a published validation-scope document. For a creator analytics tool, that framing is more useful than feature count.