Zack D Films AI Video Generator is an agent-native production pipeline for fast, curiosity-driven 3D shorts. Give your coding agent one topic and it can plan the story, build consistent character and anatomy references, render stylized keyframes, animate them, generate narration, and assemble a vertical final.mp4 with punchy edits.

It runs through the MuAPI platform (api.muapi.ai) with local Python and ffmpeg, and can be used from Claude Code, Codex, Cursor, Antigravity, or any agent that can follow SKILL.md.

Ask your agent: “Make me a Zack D Films-style short explaining what happens when you swallow gum.”

🎬 Full demo

The complete five-shot render is assembled with impact zooms and short fade, wipe, slide, and circle-open transitions:

The included demo is encoded below 10 MB so it can also be uploaded as a GitHub video attachment.

▶ Watch or download the full demo video

▶ Open the full demo video

🎥 Showcase — all beat videos

An end-to-end example: “What Happens When You Swallow Gum?”

The five source clips used by the full demo are also included in out/test_run/clips/, so each beat can be downloaded independently.

✨ Why it works

The pipeline turns a fragmented production process into one agent-guided workflow:

• Curiosity-first scripts — hooks, open loops, fast pacing, and a clear payoff.
• Consistent 3D worlds — character turnarounds and anatomy references anchor every shot.
• Signature visual language — plasticine-like materials, subsurface scattering, dramatic studio light, macro depth of field, and internal cross-sections.
• High-retention editing — push-ins on impact beats, short fade/wipe/slide transitions, voice-led pacing, music, and captions.
• Human approval at the right moments — review the beat map and visual direction before the expensive generation stages.
• Agent-native by design — the workflow, prompts, and conventions live in SKILL.md and references/.

🎨 The look

The visual target is a polished, stylized 3D educational short: expressive digital characters, smooth clay/plastic materials, high-contrast lighting, vibrant rim light, cinematic depth of field, and anatomical or mechanical cutaways that make invisible processes visible.

The style is established in the keyframe stage first. Motion is layered on afterward, so the look stays coherent from shot to shot.

🔄 How it works

One topic flows through a single beats.json project:

topic / question
  │
  ├─ 1. Trend research       Find curiosity-driven questions and hooks
  ├─ 2. Script breakdown      Write the beat map              ◀── GATE 1: approve
  ├─ 3. Character sheets      Build multi-angle 3D references
  ├─ 4. Keyframes             Render stylized 3D scene posters  ◀── GATE 2: choose look
  ├─ 5. Motion clips           Animate each keyframe with Veo 3.1
  ├─ 6. Voice narration        Synthesize the narration and BGM
  ├─ 7. Assembly               Add impact zooms, transitions, mix audio, captions
  └─ final.mp4                Vertical 9:16 short

🧩 Models and tools

🚀 Quick start

1. Install local tools

brew install ffmpeg                 # macOS
python3 -m venv .venv
source .venv/bin/activate
pip install requests pillow

On Windows, install ffmpeg and ffprobe with your preferred package manager instead.

2. Configure MuAPI

Create an API key at muapi.ai, then export it:

export MUAPI_API_KEY="sk-..."

3. Run it through an agent

Install or expose this repository to your coding agent, have it read SKILL.md, and ask for a short:

“Create a 30-second Zack D Films-style video about why mosquitoes drink blood.”

The agent should stop for approval after the beat map and again when visual direction is ready.

4. Run the stages manually

# Discover or validate a topic
python3 scripts/trend_research.py "mosquitoes"

# Create and review the script breakdown
python3 scripts/script_generator.py out/mosquitoes --topic "How Mosquitoes Drink Your Blood"

# Generate references, keyframes, motion, audio, and the final cut
python3 scripts/character_sheets.py out/mosquitoes
python3 scripts/keyframes.py out/mosquitoes
python3 scripts/clips.py out/mosquitoes
python3 scripts/audio.py out/mosquitoes --voice-sample voice_recording.mp3
python3 scripts/assemble.py out/mosquitoes

The finished project is written to out/mosquitoes/, with the final render at out/mosquitoes/final.mp4.

📁 Repository structure

SKILL.md              Full workflow for coding agents
AGENTS.md             Entry point for Codex and other agents
README.zh.md          简体中文 documentation
references/           Style, pacing, consistency, voice, and API notes
  prompt-guide.md     3D lighting, materials, and cross-section prompts
  beat-layer.md       Curiosity-loop formulas and pacing rules
  character-sheets.md Multi-angle consistency system
  voices.md           Narration and voice setup guidance
  models-and-gotchas.md
scripts/              Pipeline automation
  trend_research.py   Topic brainstorming
  script_generator.py beats.json generation
  character_sheets.py 3D turnaround references
  keyframes.py        Scene keyframes
  clips.py            Veo 3.1 motion clips
  audio.py            Narration and BGM
  assemble.py         FFmpeg assembly and effects
examples/             Ready-to-run beat maps
out/test_run/          Included clips, final demo, and keyframe showcase assets

🧠 Agent skill details

The skill is named zackd-director and is documented in SKILL.md. It is designed for requests such as:

• “Make a Zack D Films-style 3D short.”
• “Explain this medical or scientific question as a fast animated short.”
• “Turn this topic into a curiosity-loop explainer with anatomical cutaways.”
• “Create a vertical educational video with consistent 3D characters.”

For the creative rules behind the pipeline, see references/prompt-guide.md, references/beat-layer.md, and references/character-sheets.md.

⚠️ Notes

• Generation stages require a working MUAPI_API_KEY and can incur provider costs.
• Keep API keys in environment variables; never commit them to the repository.
• The output is designed for vertical short-form video (9:16).
• If a provider endpoint changes, update the centralized client in scripts/provider.py.

📜 License

MIT © 2026 Zack D Director

🔗 Related

• Vox AI Motion Graphics Generator — the paper-collage explainer counterpart.
• awesome-generative-ai-apps — more open-source AI app projects.