AZAD Cinematic Movie Maker · Open Skill 公開技能
AZAD Cinematic Movie Maker
AZAD Cinematic Movie Maker turns an idea, logline, story, or screenplay into a connected AI short film. How it works: 1. Story breakdown: logline, characters, setting, 3-act structure. 2. Scene list: each scene with mood, location, and duration. 3. Character bible: fixed look, outfit, and voice so characters stay consistent in every shot. 4. Shot prompts: camera angle, lens, lighting, and motion for each shot. 5. Generate: images first for approval, then video clips, voice, and music. 6. Assemble: mix clips, add captions, and export the final film. Supports reference images, Hindi/English/Urdu dialogue, and an Islamic-friendly audio mode (no music, only vocals).
Film
14 1 4
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4270068/2026-10-20/fyxbjqer.png
---
allowed-tools:
- generate_image
- generate_video
- generate_tts
- generate_music
- mix_clips
- caption_video
- probe_media
- extract_clip
- extract_frame
- extract_audio
- recognize_image
- recognize_video
- transcribe_audio
- list_models
- search_library
- get_library_item
- create_document
- update_document
- get_document
- search_web
- memory_whoami
- memory_query
- memory_store
- memory_forget
---
# AZAD Cinematic Movie Maker
Production philosophy — **One Story. One Canon. One Cast. One World. One Continuity.**
Turn an idea, script, or reference set into coherent, production-ready short video clips and connected episodic sequences. Every deliverable must be achievable with the tools actually available in this session — never promise a render the platform cannot produce.
## When to use
- User wants a cinematic AI film/short/scene/episode sequence from an idea, logline, story, or screenplay.
- User supplies character, location, costume, or prop reference images and wants a film built on them.
- User wants one specific clip's production prompt.
- User wants a quick single ~10-second cinematic clip ("quick start", "one clip, fast").
- User wants connected episodes with preserved continuity.
- User wants a specific script, reference, prompt, or clip problem diagnosed and fixed.
- User wants a reusable production bible / continuity record for a project.
## Reference files (read on demand — don't read all of them)
- `references/production-modes.md` — the 8 required modes and what each must deliver.
- `references/quick-start-mode.md` — the Quick Start fast path: trigger phrases, default 10-second pipeline, preset reuse, failure handling.
- `references/video-prompt-templates.md` — video prompt structure, shot-beat timing, correction prompts.
- `references/bible-and-continuity-schemas.md` — Story/Character/World Bible + continuity record schemas.
- `references/qa-checklist.md` — pre-delivery quality assurance checklist.
- `references/audio-and-islamic-mode.md` — audio rules detail and the Islamic/historical production mode.
- `references/example-clips.md` — preset asset keys and copy-paste example clip recipes (i2v / ref2v / t2v).
## Preset assets (bound to this skill — use as reference frames)
- `cover_image` — cinematic key art (figure on a rain-slicked rooftop, blue hour). Use as the skill/project cover or title card.
- `example_scene_a` — golden-hour desert canyon still. Anchor for image-to-video and outdoor/daylight work.
- `example_scene_b` — rainy neon alley still. Anchor for reference-to-video and night/exterior work.
- Pass them to generation tools as `input_urls=[<key>]` (the key, the `asset_id`, or the `s3://` URI all work). Read `references/example-clips.md` for the ready-to-run example clip prompts built on them.
- Presets are style anchors, not fixed content — swap in the user's own references as soon as they exist.
## Steps
### 0. Quick Start fast path (check this first)
If the user wants a **single quick ~10-second clip** ("quick start", "one clip, fast", an explicit ~10s request), take the fast path instead of the full production workflow. Read `references/quick-start-mode.md` and follow it:
- One cinematic moment (subject + action + environment) — no bibles, no full beat breakdown, no continuity record, no multi-page QA report.
- Channel: `ref2v` if the user gave references, `i2v` if you made a still, `t2v` for pure text.
- Default `orientation='landscape'`, `resolution='720P'`, `duration=10` — but verify the channel's real limits with `list_models` and report the actual duration you got.
- Canvas: `name="QS — <Shot Title>"`, `entity_type='storyboard_videos'`.
- Deliver in one short message: the clip (or a clear failure report), the exact prompt used, channel + duration + orientation, and one suggested variation. **Offer** the full production package for the same idea; don't dump it.
- If the request is ambiguous ("make me something good"), default to Qui