Video Generation Skill (video-generation)
A complete, production-grade pipeline for transforming raw ideas and topics into finished, polished videos. Combines structured creative direction (briefing, research, scriptwriting, shot-listing) with local command-line assembly via MoneyPrinterTurbo's CLI (cli.py or configured alias).
Preflight Check 1: MoneyPrinterTurbo Availability & Command Resolution
Before running the video generation pipeline, verify that MoneyPrinterTurbo is available and resolve the execution command:
-
Check for configured alias, function, or wrapper script:
alias video 2>/dev/null || type video 2>/dev/null || which video 2>/dev/nullIf a
videoalias or wrapper exists in the shell or PATH, use it directly (e.g.video [options]). -
If no alias or command exists, dynamically locate MoneyPrinterTurbo's
cli.py:find ~ -name "cli.py" -path "*/MoneyPrinterTurbo/*" 2>/dev/null | head -n 1- From the discovered directory (e.g.
~/Applications/MoneyPrinterTurbo):- Identify the virtual environment Python interpreter (e.g.
<MoneyPrinterTurbo_DIR>/.venv/bin/python). If no.venvis present, fallback to the systempython3. - Construct the direct execution command:
<MoneyPrinterTurbo_DIR>/.venv/bin/python <MoneyPrinterTurbo_DIR>/cli.py [options]
- Identify the virtual environment Python interpreter (e.g.
- From the discovered directory (e.g.
[!IMPORTANT] If neither an alias nor MoneyPrinterTurbo's
cli.pycan be located, stop and prompt the user to install MoneyPrinterTurbo first:"MoneyPrinterTurbo is required for video assembly and rendering, but was not found on your system. Please install it first from https://github.com/harry0703/MoneyPrinterTurbo before proceeding with video generation."
Preflight Check 2: Ollama Service Status (If using Local LLM)
If script generation is configured to use a local or cloud Ollama model, verify whether the Ollama service/process is active:
systemctl is-active --quiet ollama || pgrep -x ollama > /dev/null
[!IMPORTANT] If Ollama is needed but not running, stop and prompt the user to start the service:
"The Ollama service is not currently running. Please start it by running:"
sudo systemctl start ollama(Also ensure
ollama signinhas been completed if using Ollama Cloud models).
Preflight Check 3: Dynamic Skill Discovery & Planning
Do not hardcode or assume any specific external skills are installed on the agent. Other environments or users will have different sets of skills.
Instead, dynamically inspect the environment, identify any installed skills that could assist the video pipeline, and synthesize the findings into a Video Generation Plan.
1. Locate and Inspect Available Skills
Navigate and inspect the agent's installed skills directory or configuration. Identify any skills capable of assisting with:
- Research & Ideation: Topic discovery, factual research, domain analysis, or literature lookup.
- Copywriting & Scripting: Hook formulation, creative writing, script review, or persona/voice profiling.
- Audio & Speech: Audio cleaning, transcript analysis, silence trimming, or custom TTS.
- Visuals & Assets: Code snippet rendering, UI walkthroughs, diagrams, or image generation.
- Packaging & SEO: Title/thumbnail evaluation, search metadata, or video chapter generation.
2. User Consultation Protocol (Before Planning)
Before creating the Video Generation Plan, present any discovered relevant skills to the user to get their opinion:
- Ask a simple Yes/No:
"I found the installed skill
<skill-name>which can help with [specific task, e.g., hook scoring / audio cleaning / chaptering]. Would you like me to use it for this video? (Yes / No)" - If Yes: Mark the skill as approved to be included in the generation plan.
- If No:
Ask why:
"Understood. Could you share why you prefer not to use it?"
- Why this matters: Asking "why not" uncovers vital creative constraints and sharpens the core objective (e.g., "I already wrote the exact script", "I want a serious documentary tone, not fast-paced hooks", "Keep the natural pauses in my audio").
- Use their answers to gain direct clarity on their goals before drafting the plan.
3. Formulate the Video Generation Plan
Once the user consultation is complete, formulate a tailored Video Generation Plan based on the instructions in this skill, the user's clarified objectives/constraints, and the approved skills:
- Creative Brief: Audience, target platform, duration, aspect ratio, and tone (reflecting constraints gathered during consultation).
- Approved Skills Included: Specific tools/skills the user opted into and how they will be used.
- Script & Storyboard Approach: Hook strategy, core beats, and planned shot list structure.
- Media & Audio Sourcing: Voiceover voice (Edge TTS vs. local audio), BGM volume, and visual assets (local footage vs. stock).
- Timeline Assembly: Pacing, subtitle animation, transitions, and final render specifications.
The Topic-to-Video Pipeline
flowchart TD
A["1. Inspect Available Skills in Environment"] --> B{"2. User Consultation Protocol\nAsk opinion on candidate skills (Yes / No)"}
B -- "No" --> C["Ask 'Why not?'\n(Gives clarity on objective & constraints)"]
C --> D["3. Formulate Video Generation Plan\n(Built from instructions + clarified objectives)"]
B -- "Yes" --> E["Approve Candidate Skills"]
E --> D
D --> F["4. Research & Scriptwriting"]
F --> G["5. Shot List (Linchpin Storyboard)"]
G --> H["6. Voiceover Synthesis (TTS / Audio)"]
H --> I["7. Visuals per Scene (Stock / Local / Motion)"]
I --> J["8. Timeline Assembly (MoneyPrinterTurbo CLI)"]
J --> K["9. QC, Metadata & Final MP4 Delivery"]
Step-by-Step Pipeline Instructions
1. Clarify the Brief
Before generating code or media, pin down the creative parameters. If not explicitly provided by the user, infer sensible defaults based on the topic and platform:
- Audience: Who is watching? (General public, students, developers, buyers)
- Voice & Tone Profile: Check if the user has provided a style guide, tone preferences, or persona constraints.
- Target Duration:
- Short-form: 15s–60s (TikTok, IG Reels, YouTube Shorts)
- Mid-form: 1–3 min (Product demos, quick explainers)
- Explainer: 3–5 min (Deep-dives, tutorials)
- Aspect Ratio:
9:16(Vertical for mobile / Reels / Shorts / TikTok) — default16:9(Horizontal for YouTube / desktop / presentations)1:1(Square for Instagram / LinkedIn feed posts)
- Tone: Casual, energetic, authoritative, narrative, or humorous.
- Platform: Determines pacing, font sizes, caption styling, and hook intensity.
2. Research the Topic
Gather verified facts, compelling hooks, and accurate data points:
- Factual / Current Events: Use search tools to verify dates, statistics, proper names, and core claims.
- Evergreen / Conceptual: Structure 3–5 surprising or high-value insights.
- Identify the Hook: The single most intriguing angle to grab attention within the first 3 seconds (leveraging any research/ideation skills approved in the plan).
3. Write the Script
Draft narration formatted specifically for spoken audio, not reading:
- Rule of Thumb Pacing: ~130–150 words per minute (~2.2 to 2.5 words per second). A 30-second short needs ~65–75 words; a 60-second video needs ~130–150 words.
- Hook Validation: Stop the scroll immediately in the first 3 seconds. If any hook evaluation or copywriting tools were approved by the user, run candidate hooks through them before finalizing.
- Structure:
- Hook (0–3s): Stop the scroll immediately with a bold question or surprising statement.
- Body (3–5 beats): Progress logically through core points. Use punchy sentences without dense dependent clauses.
- Close / CTA (final 3–5s): Clear takeaway or call-to-action.
- Avoid markdown formatting, emojis, parentheticals, or citations inside the spoken narration text.
4. Break the Script into a Shot List (The Linchpin)
The shot list bridges the script with visual and audio generation. Split the script into scenes (typically 3–6 seconds each).
[!IMPORTANT] The Shot List is the most critical artifact. When scene text, visual descriptions, and timing are well-structured, downstream generation and assembly become seamless and deterministic.
Shot List Template:
| Scene # | Narration Line | Visual Description / Search Terms | On-Screen Text | Est. Duration |
|---|---|---|---|---|
| 1 (Hook) | "Deep beneath the ocean, creatures create their own light." | Bioluminescent jellyfish glowing in pitch black deep sea | DEEP SEA SECRETS | 4s |
| 2 (Beat 1) | "Over 75% of deep-sea animals produce bioluminescence." | Glowing anglerfish lure moving in the dark | 75% OF CREATURES | 4s |
| 3 (Beat 2) | "They use chemical reactions to communicate, hunt, and camouflage." | Macro shot of glowing blue photophores pulsing on squid | NATURAL GLOW | 5s |
| 4 (Close) | "Follow to uncover more secrets hidden in the abyss." | Submersible light illuminating underwater trench cliff | EXPLORE MORE | 3s |
5. Generate the Voiceover
Synthesize speech from the narration.
- Timing Authority Rule: The actual duration of generated audio files (not the estimated duration from step 4) becomes the definitive timeline anchor for scene pacing and visual cuts.
- In unified execution, MoneyPrinterTurbo handles TTS automatically via Edge TTS (free, high quality).
- For custom voiceovers or pre-recorded audio, clean the audio (using any approved audio cleaning skills) and supply
--custom-audio-file <path>.
6. Source or Generate Visuals per Scene
Match visuals to the exact duration of each scene's audio:
- Stock Footage: Pexels, Pixabay, or Coverr clips matched to scene keywords.
- Local Media: Video clips or images provided via
--video-source local --video-materials "clip1.mp4,clip2.mp4". - Image Motion: When stills are used, apply dynamic motion (
--image-motion zoom) to maintain viewer engagement.
7. Assemble the Timeline
Stitch audio, video, subtitles, and music together:
- Concatenate footage according to scene order (
--video-concat-mode sequentialorrandom). - Overlay animated word-synced subtitles (
--subtitle-display-mode word_by_word). - Mix background music (BGM) under the narration at an unobtrusive level (
--bgm-volume 0.15–0.2). - Apply scene transitions (
--video-transition-mode fade-in,slide-in, etc.).
8. Render, Review, and Quality Check (QC)
Run a QC inspection before delivering the final video:
- Timing & Sync: Does the visual change align with the voiceover topic?
- Legibility: Are subtitles readable against the background? (Checked font size, contrast, stroke).
- Audio Mix: Is the voice clear over the background music?
- Safe Zones: In
9:16vertical videos, are UI overlays (like TikTok/Reels buttons) clear of critical text? - Companion Metadata (Optional): If requested or approved in the plan, generate companion upload metadata (e.g. titles, tags, descriptions, or chapter timestamps).
- Project Delivery: Transfer the final video and assets into
<project_dir>/video-<short-title>-<datetime>/.
Output Storage & Project Delivery
Always organize and deliver the final generated video and its assets into the user's project directory rather than leaving them buried inside the internal engine storage:
1. Project Directory Resolution
- Check for Active Project / Workspace Directory:
- Check if an active workspace or project directory is present.
- Fallback Prompt if No Project Directory Exists:
- If no project directory or workspace is set/known, prompt the user for the destination path before finalizing delivery:
"No active project or workspace directory detected. Where would you like to save the generated video? (Please provide a destination folder path)"
- If no project directory or workspace is set/known, prompt the user for the destination path before finalizing delivery:
2. Output Folder Naming Format
Within the resolved project directory, create a dedicated folder named:
video-<short-title>-<datetime>/
<short-title>: A slugified 2–4 word title (e.g.deep-sea-creatures,quantum-computing).<datetime>: Current timestamp formatted asYYYYMMDD-HHMMSS(e.g.20261113-171530).- Example folder:
video-deep-sea-creatures-20261113-171530/
3. Asset Delivery & Organization
When MoneyPrinterTurbo finishes rendering in internal task storage:
- Create the timestamped folder in the project directory:
mkdir -p "<project_dir>/video-<short-title>-<datetime>" - Copy the final video and accompanying generated assets from the engine's task directory (
storage/tasks/<task_id>/) into the project folder:- Final video:
final-1.mp4$\rightarrow$<project_dir>/video-<short-title>-<datetime>/<short-title>.mp4 - Subtitles:
subtitle.srt$\rightarrow$<project_dir>/video-<short-title>-<datetime>/subtitles.srt - Voiceover audio:
audio.mp3$\rightarrow$<project_dir>/video-<short-title>-<datetime>/audio.mp3 - Script:
script.json$\rightarrow$<project_dir>/video-<short-title>-<datetime>/script.json
- Final video:
- Provide the user with the direct, clickable file link to the final
.mp4inside their project directory.
Command Execution & Task Outputs
The pipeline is driven by MoneyPrinterTurbo's CLI (cli.py or configured alias):
- Command Invocation:
Run via detected alias/command (if present) or directly using the virtual environment Python interpreter with
cli.py:# If alias or wrapper exists: video [options] # Direct execution via python: <MoneyPrinterTurbo_DIR>/.venv/bin/python <MoneyPrinterTurbo_DIR>/cli.py [options] - Configuration: Configured via
config.tomlinside the MoneyPrinterTurbo directory (contains API keys, TTS defaults, and LLM providers). - Task Output Structure: Task outputs are written inside MoneyPrinterTurbo's
storage/tasks/<task_id>/:- Final video:
storage/tasks/<task_id>/final-1.mp4 - Voiceover track:
storage/tasks/<task_id>/audio.mp3 - Subtitle file:
storage/tasks/<task_id>/subtitle.srt - Script data:
storage/tasks/<task_id>/script.json
- Final video:
- Bundled Resources:
- Fonts: Stored in
resource/fonts/(e.g.STHeitiMedium.ttc,MicrosoftYaHeiNormal.ttc,BeVietnamPro-Bold.ttf,Charm-Bold.ttf). - Music: Stored in
resource/songs/(e.g.output000.mp3tooutput029.mp3).
- Fonts: Stored in
Execution Modes & Recipes
(Note: In the recipes below, replace python cli.py with your resolved command: either your configured video alias or <MoneyPrinterTurbo_DIR>/.venv/bin/python <MoneyPrinterTurbo_DIR>/cli.py)
Mode A: End-to-End Orchestration (Recommended)
The agent drafts the script, plans the shot list, and executes the complete video render in one command:
python cli.py \
--video-script "Deep beneath the ocean, creatures create their own light. Over 75 percent of deep-sea animals produce bioluminescence to hunt, mate, and survive in eternal darkness. Subscribe to dive deeper into the abyss." \
--video-terms "bioluminescence,deep sea jellyfish,ocean abyss,underwater glowing" \
--video-aspect 9:16 \
--voice-name en-US-ChristopherNeural \
--voice-rate 1.05 \
--subtitle-display-mode word_by_word \
--subtitle-animation pop_spring \
--bgm-type random \
--bgm-volume 0.15 \
--video-source pexels
Mode B: Staged / Phased Execution (--stop-at)
Use --stop-at to inspect or iterate on intermediate assets without rendering the full video:
- Generate or test script only:
python cli.py --video-subject "Origin of Coffee" --stop-at script - Generate and inspect voiceover audio:
python cli.py --video-script "Your script here" --voice-name en-US-JennyNeural --stop-at audio - Generate subtitles:
python cli.py --video-script "Your script here" --stop-at subtitle - Download / inspect footage:
python cli.py --video-script "Your script here" --video-source pexels --stop-at materials
Mode C: Assembling Local Assets & Screen Recordings
When the user has recorded clips, slides, or local images:
python cli.py \
--video-script "Here is a quick overview of our new feature release. First, notice the updated dashboard. Next, review the live analytics." \
--video-source local \
--video-materials "./slide1.png,./screen_recording.mp4,./slide2.png" \
--image-motion zoom \
--video-aspect 16:9 \
--video-transition-mode fade-in \
--voice-name en-US-GuyNeural
Common Edge TTS Voices Catalog
| Voice Name | Language / Locale | Gender | Style / Best For |
|---|---|---|---|
en-US-ChristopherNeural | English (US) | Male | Authoritative, energetic, explainers |
en-US-JennyNeural | English (US) | Female | Warm, conversational, tutorials |
en-US-GuyNeural | English (US) | Male | Natural, casual storytelling |
en-US-AriaNeural | English (US) | Female | Professional, news, announcements |
en-GB-SoniaNeural | English (UK) | Female | Elegant, documentary, British accent |
en-GB-RyanNeural | English (UK) | Male | Documentary, crisp, polished |
es-ES-AlvaroNeural | Spanish (Spain) | Male | Standard Spanish narration |
zh-CN-XiaoxiaoNeural-Female | Chinese (Mandarin) | Female | Expressive, default Chinese voice |
no-voice | None | N/A | Silent video (subtitles & music only) |
Quick Reference CLI Options
| Flag | Values / Syntax | Purpose |
|---|---|---|
--video-script | "text" | Narration script (skips LLM generation) |
--video-subject | "topic" | Topic prompt for LLM script generation |
--video-terms | "term1,term2" | Search terms for stock footage |
--video-aspect | 9:16, 16:9, 1:1 | Aspect ratio |
--video-source | pexels, pixabay, coverr, local | Source for visual clips |
--video-materials | Comma-separated paths | Files used when --video-source local |
--voice-name | Voice identifier or no-voice | Edge TTS voice |
--voice-rate | Float (0.8 to 1.5) | Speech pacing multiplier |
--voice-volume | Float (0.5 to 2.0) | Voiceover audio gain |
--bgm-type | random, none, custom | Background music mode |
--bgm-file | Filename in resource/songs/ | Specific background music track |
--bgm-volume | Float (0.1 to 0.3) | BGM audio level |
--subtitle-display-mode | word_by_word, sentence | Subtitle animation timing |
--subtitle-position | bottom, two_thirds_bottom, center | Vertical placement of captions |
--video-clip-duration | Integer seconds (default: 5) | Max length before cutting to next clip |
--video-transition-mode | none, fade-in, fade-out, shuffle | Transition between scenes |
--stop-at | script, audio, subtitle, materials, video | Stop execution after specified phase |
Output Validation & Delivery
When execution completes successfully:
- The tool prints a JSON summary to stdout:
{"task_id": "...", "result": {"videos": ["<output_path>/storage/tasks/<task_id>/final-1.mp4"]}} - Verify the generated MP4 file exists in internal task storage.
- Create
<project_dir>/video-<short-title>-<datetime>/and transfer the final video and assets into it. - Report the absolute clickable file link of the finalized video in the project directory to the user.
- If an iteration is required (e.g. adjust speech rate or swap a term), rerun with the adjusted flags rather than rebuilding from scratch.