HeyChengC/wallstreet-explainer

What is it?

wallstreet-explainer is a "Skill" — a packaged set of instructions and tools — that you give to an AI assistant (like Claude Code) to automatically produce short, high-end financial-style videos. The style mimics professional finance news segments: clean, minimal, with real footage, big numbers, and smooth camera moves.

What does it do?

The Skill guides the AI through the entire video production process:

• Finds and uses real footage — from press conferences, product launches, official corporate videos, and news clips. It prioritizes authentic, high-quality, official sources.
• Extracts people from video — it can cut a person out of a news clip frame-by-frame and place them in front of large numbers or graphics.
• Visualizes data and relationships — it decides how to display numbers, charts, and connections in a clean, impactful way.
• Plans camera moves and transitions — it enforces a "restrained" style: no flashy PPT-like fades, no unnecessary cuts. Instead, it uses smooth, continuous moves (e.g., sliding along a table, pushing in on a chart).
• Handles voiceover and audio — it can generate a temporary voiceover using a free tool, then later swap in a real recorded voice and re-sync the video.
• Renders the final video — it draws every frame programmatically and mixes in background music, sound effects, and the voiceover.

How does it work?

The Skill is a set of files that the AI reads and follows. It includes:

• SKILL.md — the main entry point for the AI.
• references/ — detailed guides on thinking per shot, camera motion, style, grammar, opening hooks, sourcing footage, audio, review checklists, and lessons learned.
• scripts/ — ready-made tools for specific tasks:
  • Searching and downloading footage (with options to grab only a segment)
  • Capturing news screenshots and locating keywords for highlighting
  • Generating temporary voiceover and aligning timings
  • Removing backgrounds from video frames (with optional GPU acceleration)
  • Cleaning up color fringes from green/blue screens
  • Creating camera moves, depth-of-field, motion blur, and big text effects
  • Rendering the final video frame-by-frame
  • Checking if a shot is too static
  • Mixing all audio tracks together
• starter/ — a blank template for a new video (no shots, just the base structure).
• examples/ai-burn-money/ — a complete sample video (77 seconds) with its script, storyboard, code, and explanations for each shot.

How do you use it?

1. Give the Skill to your AI tool — either by sending the link to the repository or cloning it manually.
2. Install dependencies — the README lists commands for Python packages, Node.js, Playwright (for browser automation), and FFmpeg.
3. Ask the AI to make a video — for example: "Use wallstreet-explainer to make a 1-minute video about how much money AI is burning."
4. Choose a hook — the AI will propose several attention-grabbing openings; you can pick one or describe your own.
5. Optional: replace the temporary voice — after the timeline is set, you can record your own voiceover, save each sentence as a file, and run a script to re-sync the video.

Style rules (important to the creator)

The Skill enforces a specific visual language:

• Restraint — no unnecessary cuts or camera changes. If an object is the focus, keep the camera moving smoothly on it.
• No PPT-style transitions — no fades or wipes that look like slideshows.
• Real footage over graphics — use actual news clips, product launches, and official images. Highlight key numbers and quotes directly on the footage.
• Typography matters — specific fonts are recommended (e.g., MiSans for subtitles, Bebas Neue for large numbers). The user must download them separately.
• Every shot must answer four questions — Does it add new information? Does the camera move logically? Is the footage real, high-quality, and official? Are the numbers clearly explained?
• Before delivery, watch it muted — the visuals alone should tell the story.

Example included

The repository contains a full sample video: "AI 到底有多烧钱" (How much money does AI burn). It's 77 seconds long and demonstrates the style. The README shows a GIF of it (without sound). All third-party footage, people, and screenshots are credited in a separate file.

Licensing and credits

The code and documentation are MIT-licensed. However, the third-party video clips, screenshots, logos, and people shown in the sample are not part of the MIT license — they belong to their respective owners. The README advises users to include proper credits in their own videos and to be aware of legal risks.

Contact

WeChat: Yangdavee (with a note about your purpose).

Additional notes

• The Skill relies on tools like yt-dlp and edge-tts that occasionally break; the README suggests updating them with pip install -U.
• For background removal, you need onnxruntime (or onnxruntime-gpu for NVIDIA GPUs) and a model download.
• If Playwright is installed elsewhere, you can point to it with an environment variable.

---

This explainer is based solely on the README provided. No features or facts have been added beyond what is documented there.