Reborn — Master Prompt Generator

A meta-skill that turns any niche + reference YouTube video into a reusable master prompt for AI video generation. Paste your NICHE and a YOUTUBE_URL. Get back a complete, niche-locked master prompt — ready to use for every topic in that niche, forever.

Status

Last reviewed: October 2026 · release v2026.10

What this is

A niche-agnostic prompt-generation system. Where the original reborn-history repo produced one hardcoded master prompt for grimdark history videos, this repo produces a new master prompt for any niche you point it at.

You give it:

NICHE: abandoned mansion restoration
YOUTUBE_URL: https://youtube.com/watch?v=xxxxx

It runs a 5-agent reverse-engineering pipeline internally and returns:

• A complete master prompt with 33 sections — priority order, variation engine, stage progression, audio system, lighting system, scene JSON schema, anti-fail rules, style lock — all niche-locked.
• That single master prompt then works for every topic within that niche. You never need to regenerate it for a new video — just feed it new TITLEs.

Quick start

1. Open META-PROMPT.md. Copy the whole file.

2. Paste it into Claude / GPT-5 / Gemini 2.5 Pro.

3. At the bottom of the meta-prompt, fill in your INPUT block:

NICHE: luxury watch restoration
YOUTUBE_URL: https://youtube.com/watch?v=xxxxx
TARGET_TOOL: VEO3
DURATION: longform
LANGUAGE: English
INCLUDE_SCRIPT: no
EXTRA_NOTES: ASMR-tier, no narration first 8s

INCLUDE_SCRIPT is required — yes if your niche needs voiceover narration (history, lore, mystery, educational), no if it's purely visual (restoration, craft, ASMR, aesthetic loops). If you skip it, the generator will ask before producing anything.

4. Send. Receive your master prompt.

The model will silently run all 5 phases (Niche Intelligence → Video Analysis → Pattern Engine → Prompt Architect → Output Compiler) and emit the final master prompt only.

5. Save the master prompt. Use it for every topic in that niche.

For each new video, just paste the master prompt + a TITLE + (optional) reference image. The master prompt does the rest.

How it works — the 5-agent pipeline

You can run any agent standalone for diagnostics — useful for niche validation, competitive analysis, or auditing existing prompts.

What the output looks like

Every generated master prompt conforms to the structure in templates/MASTER-PROMPT-TEMPLATE.md:

• PRIMARY OBJECTIVE (PRIORITY ORDER) — conflict resolution
• INPUT SYSTEM — [TITLE] + [REFERENCE IMAGE]
• DESIGN AUTONOMY RULE — what the system never asks the user
• HIGH RETENTION SYSTEM — section-level engagement map
• VISUAL HOOK RULE — contrast moments
• VARIATION ENGINE — 4+ axes × 5+ variants
• SECTION GENERATION SYSTEM — order and logic
• STRUCTURE CONSISTENCY — what is locked
• SPATIAL / TEMPORAL CONTINUITY
• CAMERA SYSTEM
• CONTINUITY RULE / NO REGRESSION RULE / OBJECT PERSISTENCE
• {NICHE_PROCESS} STAGES — 6–10 niche-specific stages
• MICRO-STAGE DETAIL SYSTEM
• STAGE RULES / PROGRESS RULE / SYMMETRY RULE
• {NICHE_KEY_MOMENT} RULE — signature rule (e.g., "epoxy only at stage 6")
• TRANSITION SYSTEM / PACING
• AUDIO SYSTEM — per-stage SFX + music + VO
• SCRIPT WRITING SYSTEM — narration mode + register lock + hook formula + retention mechanics + cue notation + closing formula + script anti-fails + worked output example
• LIGHTING SYSTEM — per-stage progression
• COLOR GRADING LOCK
• IMAGE QUALITY SYSTEM
• MASTER IMAGE TEMPLATE — includes Character Lock: field
• CHARACTER / SUBJECT CONTINUITY — tool-specific lock method (Midjourney --cref, Flux LoRA, Kling first-frame, Runway @ref, Sora storyboard cards, VEO image_condition)
• SCRIPT-TO-SCENES PIPELINE (only if INCLUDE_SCRIPT=yes) — 7-step math: emit script → ask for VO length → seconds = min × 60 → scenes = floor(seconds/8) → images = scenes+1 → chain pairing → distribute proportionally
• SCENE JSON SYSTEM — valid placeholder schema + worked example
• OUTPUT CONTROL / FAILSAFE / STYLE LOCK

About INCLUDE_SCRIPT

This required field tells the generator which workflow your niche needs:

If unsure, the generator will ask. Niche-family default suggestions are documented in the META-PROMPT.

About the Script-to-Scenes math

When INCLUDE_SCRIPT=yes, the generated master prompt enforces this exact flow at runtime:

1. Write script (text only)
2. Ask user: "VO length in minutes?"
3. TOTAL_SECONDS = MINUTES × 60        # e.g., 8 × 60 = 480
4. SCENE_COUNT = floor(SECONDS / 8)    # e.g., floor(480/8) = 60
5. IMAGE_COUNT = SCENE_COUNT + 1       # e.g., 60+1 = 61

Scene 1 = Image 1 + Image 2
Scene 2 = Image 2 + Image 3
...
Scene 60 = Image 60 + Image 61

The same image acts as both the END of one scene and the START of the next — no jump cuts. (Sora 2 special case: IMAGE_COUNT = SCENE_COUNT, no end-frame.)

About the Script Writing System

The Script section is always present, but its MODE adapts to the niche:

• full-narration — continuous VO, locked register, paced wpm. Used for history, lore, mystery, educational, "what happened to" niches.
• minimal-narration — short whispered VO under 30 wpm averaged, mostly silence. Used for high-end ASMR-craft.
• text-only — on-screen text cards, zero VO. Used for aesthetic / cinemagraph / pure-visual niches.
• silent — no VO, no text — pure SFX + music. Used for loops and abstract aesthetic niches.

Even in silent mode the section is still emitted (with VO explicitly forbidden), so the downstream model never has ambiguity about what is and isn't allowed.

About Character / Subject Continuity

The Character Continuity section is always present, but its method adapts to the niche and target tool:

• History / lore / mystery with named figures → midjourney_cref (default) or flux_lora (for high-volume channels)
• Restoration / craft with anonymous hands → none
• Documentary with reference photos available → runway_references
• Long-form storytelling with consistent characters → sora_storyboard
• Pure motion / cinemagraph → none

When method ≠ none, every still-image prompt's Character Lock: field carries the actual tool syntax (e.g. --cref https://cdn.midjourney.com/<hash>/0_0.png --cw 100), so downstream models can paste-and-go.

About Numbering Consistency

Every master prompt picks ONE numbering scheme and applies it everywhere:

• Scheme A — Section-aligned (1-indexed): "Section 1 / Stage 1 — Cold Open" through "Section 8 / Stage 8 — Coda". Used for fixed-arc niches (history, lore).
• Scheme B — Process-aligned (0-indexed): "Stage 0 — Abandoned" through "Stage 10 — Final Reveal". Used for transformation niches (restoration, craft).

The HIGH RETENTION SYSTEM, STAGES, PACING, AUDIO SYSTEM, LIGHTING SYSTEM, and SCRIPT WRITING SYSTEM all use the same numbering. No mixed indexing — this was a v2.1.0 bug, fixed in v2.2.0.

See examples/ for sample inputs and the master prompts they produced.

Repository layout

reborn-master-prompt-generator/
├── README.md                                   ← you are here
├── SKILL.md                                    ← Claude skill manifest
├── META-PROMPT.md                              ← THE generator (paste-and-use)
├── LICENSE                                     ← MIT
├── CHANGELOG.md
├── CONTRIBUTING.md
├── MIGRATION-GUIDE.md                          ← how to upgrade from reborn-history v1
├── agents/                                     ← 5-agent pipeline (modular)
│   ├── 01-niche-intelligence.md
│   ├── 02-video-analyzer.md
│   ├── 03-pattern-engine.md
│   ├── 04-prompt-architect.md
│   └── 05-output-compiler.md
├── templates/
│   └── MASTER-PROMPT-TEMPLATE.md               ← canonical output skeleton
├── examples/
│   ├── input-mansion-restoration.md
│   ├── input-grimdark-history.md
│   └── input-luxury-watch-restoration.md
├── prompts/
│   └── legacy/                                 ← original reborn-history master prompts
│       ├── MASTER-PROMPT-REBORN-ENGLISH.md
│       ├── MASTER-PROMPT-REBORN-HINDI.md
│       ├── MASTER-PROMPT-LONG-HINDI.md
│       └── MASTER-PROMPT-SHORT-HINDI.md
├── reference/                                  ← original 5-agent outputs (history niche)
│   ├── 01-niche-intelligence.md
│   ├── 02-video-analysis.md
│   ├── 03-pattern-engine.md
│   ├── 04-prompt-architect.md
│   ├── 05-output-compiler.md
│   ├── tim-reborn-research.md
│   └── style-guide.md
├── tools/
│   └── _build.py                               ← markdown → PDF builder
└── .github/

Two-step workflow

Step 1 — Generate (do this once per niche):
META-PROMPT + NICHE + YOUTUBE_URL → master prompt for that niche.

Step 2 — Produce (do this for every video):
master prompt + TITLE + (optional) REFERENCE IMAGE → script + image prompts + scene JSONs.

The master prompt produced in Step 1 is reusable. You only run the generator once per niche, then feed it to your image/video model with new topics for every video you make.

Example niches this works for

• Abandoned mansion / hotel / castle restoration
• Cinematic AI history (grimdark, golden-hour, mystery, war)
• Ancient ruins exploration / civilization reconstruction
• Luxury car / watch / jewelry detailing and restoration
• Satisfying ASMR-craft (knife forging, leather work, woodworking)
• Mythology and folklore retellings
• "What happened to" mystery / lore breakdowns
• Disaster cinematics (volcanoes, tsunamis, fires)
• Aircraft / ship / weapon technical history
• Travel "what it's like to live in" cultural cinematics

If your niche has a clear visual + audio signature and a repeatable section flow, the generator will produce a good master prompt for it.

Languages

The generator supports any language. Just pass LANGUAGE: <language> in the INPUT block. The output master prompt will lock register, vocabulary, and pacing to that language.

For non-English niches, supply a reference video in that language for best results — the Video Analyzer agent will pick up native pacing and register cues.

Stack assumed by generated prompts

The generated master prompts default to the Midjourney v7 + Kling 2.5 + ElevenLabs v3 + VEO 3.1 stack, but you can override per niche with TARGET_TOOL in the INPUT block. Supported targets:

Contributing

PRs welcome — especially:

• New example niches (input + the master prompt it produced)
• Locale-specific register notes (vocabulary tables, pacing rules)
• Tool-specific JSON schema additions (Sora 2, Wan 2.5, Hailuo)
• Improvements to the agent prompts (better failure detection, sharper variants)

See CONTRIBUTING.md.

Migration from reborn-history v1

If you were using the original reborn-history (history-niche-only) version, see MIGRATION-GUIDE.md. The 4 original master prompts (English, Hindi long, Hindi short, Hindi grimdark) are preserved in prompts/legacy/ and continue to work standalone — they are now examples of what the generator produces.

Credits

• Architecture parent: reborn-history (the history-niche-specific predecessor)
• Reference channel for the original: Tim - Reborn History
• Built by: Waseem Nasir / SkynetLabs

License

MIT — free for commercial use. Attribution appreciated, not required.

Work with SkynetLabs

Built by Waseem Nasir at SkynetLabs — n8n + AI automation for small businesses.

• Hire on Fiverr: https://www.fiverr.com/agencies/skynetjoellc
• Hire on Upwork: https://www.upwork.com/freelancers/muhammadwaseemnasir
• Book a free call: https://calendly.com/skynetlabs/schedule-a-free-consultation