HeyChengC/wallstreet-explainer

This is a production kit, not a one-click generator: it packages a style guide, a set of sourcing/rendering scripts and a sample film so an AI coding agent can help you build a tightly styled finance explainer short from real footage and code-drawn graphics. If you want a repeatable editorial look rather than a generic AI slideshow, it is worth examining; if you need a turnkey video tool, it will be too hands-on.

The problem

Making a finance or tech explainer short with real footage usually means hunting press conferences and product launches, cutting a usable clip, cutting a person out frame by frame, placing a number and a chart over it, then recording a voice-over and mixing music underneath. Each step lives in a different tool, and the visual style drifts as you go. Here the whole chain is written down as a skill and a set of scripts: the agent is told how to source, how each shot should move and what to check before delivery, and the finished timeline is rendered frame by frame with code. The result is a video whose look is governed by the same rules from the first shot to the last, plus a sample film that shows what those rules produce.

What's different

The distinctive idea is treating shot design as code. The repo ships a starter base that reads a timeline, registers shots and subtitles, plus a moves library with worked actions such as sliding between people and swapping the background behind a cutout, leaving the actual composition of each film to be written per project. Supporting scripts handle the physical chores: searching and downloading footage, building contact sheets, extracting frames, capturing news pages and measuring keyword positions for highlights, matting people out of video, removing colour spill at the matte edge, rendering, spotting motion that looks too static, and mixing dialogue, original sound, effects and music. Delivery is checked against a review list, including watching the film muted to confirm the picture alone carries the story.

The trade-off is that this is a kit of parts with heavy dependencies, not an application. The author's style rules are also prescriptive — restraint over cuts, few small labels, no PPT-style fades — so a project that wants a different editorial voice will be working against the guide rather than with it. And the sample film is evidence of one produced result, not an independent benchmark of output quality.

Who should try it

A good fit is a solo creator or small team already comfortable with a coding agent, willing to supply or approve a script and voice-over, and wanting a consistent finance-style look across episodes rather than hand-editing each one in a timeline app. It also suits someone who wants to learn the shot grammar from the references and the annotated sample rather than copy a template. A poor fit is anyone expecting to type a topic and receive a finished video: the README describes choosing a hook, a script, footage and final narration as part of the workflow, and the scripts assume you can install and troubleshoot the toolchain. The README's main path is a Claude Code skill install, so another AI tool would need its own route to load the skill.

Requirements and limitations

Budget for setup. You need ffmpeg on PATH, Python requirements and Node with a Playwright Chromium install; a coding agent such as Claude Code is the documented entry point, with the repository cloned into its skills directory. Person cutouts need ONNX Runtime (GPU build if you have an NVIDIA card) and a rembg model; switching from the free temporary voice to your own recording requires faster-whisper for word-level realignment, again much faster with an NVIDIA GPU. Fonts are not bundled: MiSans for subtitles, Alibaba Momo Number Black for labels and titles, Bebas Neue for large figures, Montserrat for small English labels and Noto Sans SC as fallback, with filenames matching the CSS; one subtitle italic font is commercial and the README suggests substitutes.

On the footage, the project is explicit that it uses third-party corporate, news and interview material for commentary, with credits recorded per film, and leaves risk judgement to the user; the code and docs are MIT but the sample's third-party frames, screenshots, logos and people are not. Upkeep is real: the README warns that yt-dlp and edge-tts can start returning 403 and need updating, and Playwright installed elsewhere must be pointed to via an environment variable.

HotRepo's take

HotRepo first recorded this repository on the 2026-10-26 new-repository board for JavaScript at #22, with 400 total stars; that is a board rank, not a daily growth figure, and the stored sample covers a single date. With a project age of 7 days and no comparable daily history, persistence of interest cannot yet be judged from these signals. Before investing setup time, check that the non-bundled fonts and the documented dependency versions line up with your machine and your agent, since those are the first places a first render will stall. For a style-driven workflow and code-drawn graphics, the references and sample are the parts worth reading first.

Project source and documentation