Erduo Broll Loop Engineering: SRT-Driven B-Roll Generation for Agentic Video

Where this fits

You have a client's raw transcript (SRT) and a design brief. Instead of manually storyboarding, sourcing stock footage, and animating every shot, you hand both files to an AI agent. Erduo Broll Loop Engineering turns that SRT into a structured shot list, generates source materials, creates animations, and directly renders each shot—ending with a full preview. It's built for developers who want the agent to own the video production loop, from semantic breakdown to final master render.

Why agents benefit

• SRT-driven storyboarding: The agent parses the transcript's timing and dialogue to output original shot cards (152 covered), skipping the manual cut-by-cut design phase.
• Dual-backend routing: Automatically chooses between HyperFrames and Remotion based on shot complexity, resolution needs, or motion requirements—no config scripts to maintain.
• Canary approval workflow: Renders low-res previews of key shots first; the agent waits for your sign-off before committing to a full-resolution master, saving GPU hours.
• Loop-based iteration: Treats the entire pipeline—storyboard, asset fetching, animation, rendering—as a repeatable loop, so redoing a shot isn't a full restart.
• Environment-aware setup: Runs a deep setup check on installation (macOS, Codex/Claude Code) and flags missing dependencies before any render, reducing "works on my machine" failures.

Practical scenarios

Batch podcast clip production
A content team has 20 interview SRTs. The agent ingests each file, generates a distinct b-roll plan per episode, renders previews, and only after human approval produces the 4K masters for all episodes in one run.

Client pitch with tight deadline
You have a 48-hour turnaround. You drop in the script and a reference design. The agent produces a 40-second sample with original visuals overnight. You review the canary shots, request two color changes, and approve—the final render is ready by morning.

Localized video for multiple markets
With the SRT in multiple languages, the agent reuses the same shot cards but adapts text overlays and pacing per language, then renders localized masters without redoing the whole storyboard.

Add it to your agent workflow

Install directly from the skill registry (preferred) or GitHub:

npx skills add erduo1998-cell/erduo-broll-loop-engineering

Usage example in your agent prompt (Codex or Claude Code):

> Install the skill and open the latest English SRT. Build a shot list, generate previews for the first 5 shots, and wait for my approval before rendering the final 4K master.

The skill's canary step will pause the agent until you explicitly confirm.

Compare before adopting

Validate against three dimensions: (1) quality signals—does the skill actually exercise both HyperFrames and Remotion, or does it only wrap one backend?; (2) maintenance freshness—check the changelog and recent commits; a video toolchain that isn't updated risks breaking with new Remotion releases; (3) workflow fit—weigh it against hand-rolled scripts or generic animation libraries if your pipeline is already tightly integrated. Also note the README explicitly avoids promising visual consistency across the two backends, so decide if that's acceptable for your use case.

Why it is worth tracking

With 71 stars in a niche category, Erduo is a quiet signal that agent-driven video is moving from demos to production—the skill's design reflects a practical understanding of render costs and human-in-the-loop approval. Track it if you're assembling a video agent toolkit; evaluate it as soon as your team needs repeatable b-roll generation from transcripts. More details on the skill page can guide your initial test.