How to use Edit Drift: what you can do with it

3 use cases

• Score how much an image model changes what it shouldn't

  Run the 40-edit benchmark with your model and compute its Edit Fidelity Score.

  • REPO MegaTroll222/edit-drift
  • Data and analytics
  • Advanced
  View prompt

  Using MegaTroll222/edit-drift, run the milk benchmark edits with <model>, score them, and report the Edit Fidelity Score and final drift.

  Share Email me this

• Compare image models with a leaderboard and video

  Generate a report and side-by-side videos for every model you scored.

  • REPO MegaTroll222/edit-drift
  • Data and analytics
  • Intermediate
  View prompt

  Run `editdrift report benchmarks/milk results/milk --out report` and summarize the ranking. Create a side-by-side video of the top two models.

  Share Email me this

• Run the benchmark from Claude Code

  Install the plugin and let Claude Code drive scoring and reporting.

  • REPO MegaTroll222/edit-drift
  • Claude Code
  • Data and analytics
  • Intermediate
  View prompt

  Using the edit-drift Claude Code plugin, score results in <folder> and produce the report. Confirm scores.json exists for each model.

  Share Email me this

Quick start with Edit Drift

1. Check you have what it needs: Python with pip; Outputs from the image models you want to compare.
2. Open MegaTroll222/edit-drift on GitHub and read the README and license.
3. Install with pip install git+https://github.com/MegaTroll222/edit-drift
4. List the edits with editdrift steps benchmarks/milk
5. Generate each edit with your model, applying it to the previous result
6. Score with editdrift score benchmarks/milk results/milk/my-model

These steps come from our top use case, Score how much an image model changes what it shouldn't, which also has a copy-paste agent prompt.

Using the Deduped MCP, get the setup steps for MegaTroll222/edit-drift

claude mcp add --transport http deduped https://dedupedai.com/mcp

Edit Drift alternatives · Best AI repos for Data and analytics

About this repo

• Who it is forML engineers and teams choosing an image editing model who care about fidelity in untouched regions.
• Why it is hot14 stars, launched 19 days ago, picked up by one AI source we track
• LanguagePython
• Sourcegithub.com/MegaTroll222/edit-drift · Back to Hot repos

Requirements

• Python with pip
• Outputs from the image models you want to compare

Related repos

• Trended earlier REPO google/langextract LangExtract LangExtract is a Google Python library that uses LLMs to extract structured information from unstructured text with precise source grounding and an interactive visualization.
• Trended earlier REPO unclecode/crawl4ai Crawl4AI Crawl4AI is an open-source web crawler and scraper that turns websites into clean, LLM-ready Markdown for RAG, agents and data pipelines, self-hosted or via a hosted API and MCP server.
• Trended earlier REPO firecrawl/firecrawl Firecrawl Firecrawl is an API that searches, scrapes, crawls and interacts with the web and returns clean markdown, structured JSON or screenshots for AI agents. It handles JavaScript-heavy pages, proxies and rate limits, parses PDFs and DOCX, can click and type before extracting, and plugs into agents as a skill or an MCP server. It is open source and also offered as a hosted service with an API key.