Professional Movie Maker · সর্বজনীন Open Skill

Professional Movie Maker

A professional-grade AI filmmaking skill designed for creators, studios, and brands aiming for the top of the feed. Built specifically to dominate high-retention platforms like TikTok, Instagram Reels, and YouTube Shorts, Movie Maker transforms simple concepts into visually stunning, cinematic micro-films. Powered by a native ref2v pipeline, it eliminates character drift and disjointed frames, delivering flawless continuity across cuts, lighting, and action. From scroll-stopping hooks to powerful endings, this skill automates the entire Hollywood workflow to give you studio-quality videos that drive multi-million-view engagement

@aslam2noor4

Film

17 4 12

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/1261060/2026-09-28/k72dspyv.jpg

# Movie Maker

A native production skill that drives an end-to-end **reference-to-video (ref2v)** pipeline for one cohesive 10-second cinematic clip. No storyboard grids, no shot-by-shot image chaining — just one short where characters, locations, and props stay consistent across motion and cuts.

## When to use
- Vertical micro-dramas and cinematic shorts.
- Character-driven scenes requiring strict identity consistency across camera cuts.
- Reference-heavy pipelines: upload your hero, location, and key prop → bind → render.
- Episodic short-form threads: Episode 1's ending button sets up Episode 2's starting hook.
- Rapid pitch bumpers, tone proofs, and proof-of-concept clips.
- Targeted iterations on script, shot beats, reference plates, orientation, or style.

## When NOT to use
- Multi-panel storyboard grids.
- Open-ended non-linear video editing (NLE).
- Multi-turn conversational memory without an explicit project/series bible.

## Modes
| Mode | You Provide | Agent Delivers |
|---|---|---|
| Full Auto | A single idea or premise sentence | Script → Plan → References → ref2v Render |
| Script-First | A complete script or rough draft | Plan → References → ref2v Render |
| Reference-Led | 2–3+ reference images + short brief | Bind uploads → Plan → ref2v Render |
| Single-Stage Edit | "Rewrite the dialogue" / "Regenerate the location ref" | Updates the targeted stage only |
| In-Thread Sequel | "Episode 2" / "Next episode" | Continuity block + asset URI reuse |

## The pipeline
| Stage | Output |
|---|---|
| Script | Logline, characters, dialogue, and shot pacing |
| Planning | World environment, visual style, key entities, and 3–6 shot targets |
| Reference Assets | Character / location / prop imagery (or user-provided uploads) |
| ref2v Prompt | Structured 6-part prompt with [Shot N] timestamps, audio cues, and dialogue |
| Render | A single 10-second video clip · 720p · Portrait or Landscape |

## Production stack (defaults)
- **Engine**: `ref2v` (reference-to-video) — single-clip ref2v, no per-shot i2v chaining.
- **Duration**: 10 seconds. Use segment + `mix_clips` for extended cuts.
- **Shots**: 3–6 timed beats, scripted directly within the ref2v prompt.
- **Orientation**: `portrait` (9:16). `landscape` available on demand.
- **Resolution**: `720P` + target aspect ratio, locked across image and video generation.
- **Series**: Hook + Outro Button; continuity maintained across conversational context.

## Steps
1. First use `write_todos` to lay out the pipeline stages, then update it as you progress.
2. Determine the mode from the user's request (Full Auto / Script-First / Reference-Led / Single-Stage Edit / In-Thread Sequel).
3. Read `references/pipeline.md` for the detailed stage-by-stage craft, the six-part ref2v prompt template, and canvas conventions — read it before producing any stage output.
4. Produce the Script (logline, characters, dialogue, shot pacing) and the World & Shot Plan (environment, visual style, key entities, 3–6 shot targets). Save these as documents with `create_document` (`entity_type='scripts'`).
5. Build Reference Assets: generate character / location / prop images with `generate_image` (`model='t2i'` for new, `model='i2i'` when editing an existing ref; `entity_type='characters'` / `'locations'` / `'props'`). User-uploaded assets take priority — bind them as references.
6. Compose the six-part ref2v prompt per the template in `references/pipeline.md`, embedding [Shot N] timestamps, audio cues, and dialogue.
7. Render with `generate_video` (`model='ref2v'`, `input_urls` = the reference images, `duration=10`, `orientation='portrait'` or as requested, `resolution='720P'`, `entity_type='final_videos'`). For dialogue, generate voiceover with `generate_tts` (`model='tts.design'` + `voice_instruction`, or `model='tts.clone'` + `reference_audio_url`); for soundtrack use `generate_music` (`model='music'`). For extended cuts, segment and stitch with `mix_clips`.
8. On success, summarize the clip in natural