مونتاج ريلز كامل ببرومت واحد باستخدام Claude Opus 5.5

You are the senior Arabic short-form editor, motion designer, and HyperFrames implementer for this task.

Inside the project folder you will receive only one raw talking-head video whose filename begins with `rawreel` (`rawreel.mp4`, `rawreel.mov`, or a similar video extension).

There is no reference video, no SRT, no storyboard, and no separate B-roll. Everything required to reproduce the requested editing style is specified below. Do not ask me to attach a reference video or subtitle file.

Your job is to inspect, transcribe, edit, design, animate, validate, preview, and finally render a premium Arabic 9:16 reel using **HyperFrames as the primary and final system**.

Do not return only recommendations, code snippets, a storyboard, or an edit decision list. Build the complete working HyperFrames project. Stop only at the final Studio-preview approval gate required before rendering.

## 1. Non-negotiable framework choice

Use **HyperFrames** for composition structure, source-range editing, captions, graphic interludes, animation, audio placement, Studio preview, validation, and rendering.

Do not replace the project with a pure Remotion implementation. Remotion may be used only as a subordinate helper if the installed HyperFrames workflow explicitly requires it. HyperFrames remains the source of truth.

Read and obey the installed skills in this order:

1. `/hyperframes`
2. `/general-video` — this is a footage remix with silence removal, hard cuts, reframes, captions, and full-screen editorial interludes.
3. `/hyperframes-core`
4. `/hyperframes-creative`
5. `/hyperframes-animation`
6. `/hyperframes-keyframes`
7. `/hyperframes-studio`
8. `/media-use`
9. `/hyperframes-audio` for dialogue processing or any audio mixing
10. `/hyperframes-registry` before hand-building a named texture, effect, or transition
11. `/hyperframes-cli`

Refresh the supported `general-video` workflow first. Create the edit in a new child folder such as `hyperframes-reel-edit/`. Never overwrite or modify the original `rawreel` file.

## 2. Inspect and understand the source before designing

Locate the single video matching `rawreel.*`, then:

1. Run `ffprobe` and record:
   - duration;
   - width and height;
   - frame rate;
   - video/audio codecs;
   - audio sample rate and channel count.
2. Confirm that the source is a vertical talking-head reel. Preserve its native frame rate whenever practical. The intended delivery canvas is 1080×1920, 9:16.
3. Generate contact sheets across the entire source at 1–2 fps so you understand framing, gestures, facial movement, available negative space, lighting, and possible crop positions.
4. Analyze the waveform and detect silence/breath intervals. Do not cut from numeric silence detection alone; verify speech and visible gestures.
5. Extract the dialogue and produce a highly accurate Egyptian/Arabic transcript with word-level timestamps.

Use the supported HyperFrames/media transcription workflow when available. Otherwise use a reliable local Whisper-compatible method. Save these inspectable artifacts inside the project:

- `transcript.json` — word-level tokens containing `text`, `start`, and `end`;
- `captions.srt` — corrected Arabic cue transcript;
- `source-analysis.md` — metadata, silence ranges, content summary, and the speaker’s major semantic beats;
- `timing-map.json` — every retained source range mapped to its new output start/end after silence removal.

Review the Arabic transcript manually against the audio. Correct model names, English technical words, and Egyptian dialect mistakes. Never animate or publish an unverified hallucinated transcript.

## 3. Required editing style — self-contained reference description

The desired result alternates between two worlds:

### World A — restrained talking head

- The presenter remains visually dominant.
- Use clean hard cuts and a small number of motivated crop changes.
- Use three crop states as guidance:
  - 1.00× medium/base;
  - approximately 1.08× emphasis;
  - approximately 1.14–1.17× close emphasis.
- Change crop when a new claim, contrast, punchline, or CTA begins—not every sentence.
- Use a hard punch between some states and a short smooth push for the opening.
- Animate an inner visual/crop wrapper, never the timed `.clip` element.
- Keep eyes, chin, and active hands safe. Derive x/y offsets from the real footage; do not assume the face is perfectly centered.
- Keep color correction natural. Do not oversharpen the face, crush the blacks, or make skin orange.

### World B — full-screen editorial motion-design interludes

These are not generic stock-footage inserts. They are designed 9:16 editorial scenes that replace the presenter for meaningful portions of the narration.

Each interlude uses:

- warm tactile paper/cream background;
- one strong monochrome symbolic object, diagram, editorial illustration, or simple UI metaphor;
- large Arabic kinetic typography with extreme but controlled hierarchy;
- one dominant keyword and smaller supporting context;
- progressive assembly synchronized to the spoken phrase;
- a clear entrance, build, proof/readable hold, and decisive exit/cut.

The central object must be chosen from the actual transcript. Examples of acceptable semantic metaphors:

- automation/process → assembled timeline, conveyor/path, cursor, connected modules;
- thinking/decision → abstract brain/network/decision tree;
- warning/problem → fractured card, downward path, barrier, redaction mark;
- result/growth → completed frame, rising route, aligned outputs, publication card;
- steps/features → stacked modules, numbered blocks, tool icons;
- comment/CTA → oversized keyword plus comment-bubble outline.

Do not reuse the same object in every scene. Do not use random robots, neon brains, generic circuit boards, sci-fi HUDs, floating particles, or unrelated stock clips.

If a visual can be built truthfully using HTML/CSS/SVG, do that. If a symbolic image/icon is required, resolve or generate it through `/media-use`, freeze it locally, and record provenance. No render-time network requests.

## 4. Pacing and A-roll/B-roll ratio

Target an overall composition of approximately:

- 45–55% talking-head footage;
- 45–55% full-screen editorial graphic interludes.

Adapt the ratio to the transcript. Do not force B-roll over an emotional or credibility-critical statement that benefits from seeing the presenter.

Typical graphic interlude duration:

- short: 1.8–2.5 s;
- normal: 2.8–4.0 s;
- longest list/process scene: no more than approximately 4.5 s unless the speech genuinely requires it.

Use roughly 4–6 semantic scenes for a 20–35 s reel. Group adjacent speech into purposeful beats; do not make one scene per sentence or subtitle line.

Required narrative rhythm:

1. Open on the presenter for approximately 0.7–1.2 s.
2. Hard-cut on the hook’s proof/claim into the first full-screen editorial interlude.
3. Return to the presenter for context or credibility.
4. Use a second graphic interlude for the mechanism, input, process, or contrast.
5. Use the longest graphic scene for a list, transformation, or sequence of benefits.
6. Return to the presenter for the payoff/result.
7. Keep the presenter visible during the CTA setup, then optionally cut to a final designed CTA card on the spoken keyword.

Never alternate mechanically every fixed number of seconds. Content meaning determines the boundary.

## 5. Opening hook treatment

The first second must feel deliberately edited.

- Start on the presenter.
- Use a fast elegant push-in from the base crop to approximately 1.10–1.13× over 0.55–0.85 s with a strong ease-out.
- Do not bounce or return to 1.00× immediately.
- Introduce the first small word-timed caption during the push.
- The first full-screen graphic cut should happen on the strongest claim or named tool/result in the hook, normally within the first 1.0–2.0 s.
- Do not start with a logo, black frame, empty background, generic title card, or slow fade.

## 6. General editorial color system

This is a personal, unbranded visual treatment. Do not add an organization name, logo, website, watermark, or invented brand identity.

Use this warm editorial palette as a flexible starting point. Small adjustments are allowed to harmonize with the source footage, but preserve the same functional hierarchy.

```css
--paper: #E9E0D3;
--paper-dark: #D4C8B8;
--paper-light: #F6F0E7;
--cream: #FFFDF7;
--ink: #1B1D1E;
--ink-soft: #3A3D3E;
--accent-warm: #E66A21;
--accent-warm-dark: #C85218;
--accent-cool: #2A8791;
--accent-cool-dark: #216A72;
--accent-gold: #D8A44A;
```

Rules:

- Paper/cream/ink form the dominant visual world.
- Burnt orange is the main emphasis color.
- Muted teal is a less frequent secondary accent.
- Muted gold is a rare micro-accent, never the main theme.
- Avoid generic neon-purple AI visuals, rainbow palettes, heavy glassmorphism, and glossy corporate templates.
- A subtle paper grain is allowed only if resolved locally through `/media-use` or a suitable registry primitive.

## 7. Font handling and Arabic typography

No custom font files will be supplied. Discover and use a good Arabic font already installed on the machine.

Preferred candidates, in order of actual availability and visual suitability:

1. `Cairo`
2. `Tajawal`
3. `Noto Kufi Arabic`
4. `Noto Sans Arabic`
5. `IBM Plex Sans Arabic`
6. `Segoe UI`
7. `Arial`

Do not assume a font is installed merely because its name appears in this list. Test candidates with `document.fonts.check()` or the operating system’s font inventory, select the first strong Arabic display family that is genuinely available, and record the selected family in the delivery report.

- Use a real heavy/bold weight for display headlines when available.
- Use bold or semibold for captions and supporting headings.
- Use medium/regular for smaller explanatory copy.
- Do not synthesize a weight that renders poorly.
- Use a CSS fallback stack beginning with the selected verified family, followed by other verified Arabic-capable system fonts.
- Keep the render local and deterministic. Do not fetch web fonts at render time.
- Confirm that the selected font loads and renders joined Arabic correctly in the actual HyperFrames preview before approval.

All Arabic must use correct RTL shaping:

- `dir="rtl"`;
- whole-word spans;
- correct Unicode shaping and punctuation;
- natural Arabic line breaks;
- no individual-letter splitting or animation;
- animate complete shaped words or lines only.

## 8. Caption system

Do not display large full-sentence subtitle blocks copied directly from the transcript.

### On talking-head sections

- Derive word timing from the verified `transcript.json`.
- Show 1–4 words at a time.
- Place captions around 64–70% of frame height, normally over the torso while staying above platform UI.
- Use the selected system Arabic font in a bold or semibold weight, with compact line height and no opaque background pill.
- Default: cream `#FFFDF8`.
- Context/prior words may fall to 45–60% opacity.
- Highlight only one genuinely important word with the warm accent; use the cool accent more rarely.
- Animate with a 2–4 frame opacity/vertical settle or a clean hard replacement.
- No bouncing karaoke, per-letter wobble, overshoot on every word, or captions covering the face.

### On full-screen graphic interludes

- Do not add a second conventional subtitle layer.
- Convert the spoken phrase into the designed typography of the scene.
- Reveal semantic chunks at the exact spoken moment.
- Preserve every important factual claim without inventing new facts.

## 9. Automatic semantic shot planning

After transcription, divide the narration into semantic beats. Create a custom plan from the actual words using this decision system:

### Beat type A — Hook or surprising claim

- Begin on the presenter.
- Move to a graphic scene when the key claim/tool/result is spoken.
- Use one oversized keyword, one short supporting phrase, and one symbolic object.

### Beat type B — Input or “what I did”

- Show the presenter for the personal statement.
- Move to a graphic process when inputs are named.
- Suitable layout: source frame/card + prompt/document + arrow/path → output.

### Beat type C — List of features/actions

- Use the longest editorial interlude.
- Build 3–5 modules one at a time.
- The active module is oversized and uses the warm accent; completed modules settle to ink, cream, or the cool accent.
- Avoid making it look like a generic SaaS dashboard.

### Beat type D — Payoff or finished result

- Return to the presenter for credibility.
- Use a medium crop or subtle punch-out to create resolution.
- Highlight the result phrase across one or two caption states.

### Beat type E — CTA

- Keep the presenter visible through the setup.
- On the spoken CTA keyword, either use a motivated punch-in or hard-cut to the final graphic card.
- Final card: oversized keyword, a minimal relevant outline icon, and one short Arabic instruction.
- Hold the completed card for 12–18 frames after the final spoken word.
- Do not end on black.

Write the resulting content-specific plan into `STORYBOARD.md`. It must quote the exact verified transcript phrases that drive each scene and include the post-trim output times.

## 10. Motion language

The reel’s energy must come from hierarchy changes, progressive assembly, and decisive cuts—not nonstop movement.

Use 2–4 intentional motion rules per graphic scene:

- masked line/word rise;
- short staggered assembly;
- scale settle;
- path or connector draw;
- small object rotation/pose change;
- restrained camera drift;
- semantic shared-element handoff.

Timing guidance at the measured project frame rate:

- entrance: typically 6–12 frames;
- small stagger gap: 2–5 frames;
- readable proof/hold: typically 10–20 frames;
- hard cuts are preferred over decorative dissolves.

Use one focal object and one dominant phrase at a time.

Avoid:

- continuous random motion;
- generic glitch unless the transcript specifically justifies disruption/error;
- liquid morphs;
- chrome 3D text;
- excessive parallax;
- fake depth made only by scaling;
- spinning logos;
- particle fields;
- transition-pack variety;
- ending animations that reset to their initial state.

## 11. Silence removal and source editing

Use the waveform, transcript, and visible mouth/gesture motion together.

- Remove leading and trailing dead air.
- For internal silence longer than approximately 220–280 ms, consider trimming it while preserving 2–4 frames of natural room.
- Do not delete every breath or micro-pause.
- Never cut inside a spoken syllable.
- Avoid cutting off a hand movement that visually completes the sentence.
- Use hard cuts between retained source ranges.
- Maintain audio/video sync exactly.

Implement source editing through separate HyperFrames media elements using:

- `data-media-start` for the source offset;
- `data-duration` for the retained source length;
- `data-start` for authored output placement.

If audio is separated, its ranges and placements must exactly match the picture edits.

Keep uninterrupted dialogue under full-screen editorial scenes.

## 12. B-roll without supplied B-roll files

Do not stop because no separate B-roll files were supplied.

Create the visual replacements from:

- HTML/CSS/SVG editorial compositions;
- simple local icons/illustrations resolved through `/media-use`;
- locally extracted stills or crop details from `rawreel` when useful;
- generic, truthful UI metaphors derived from the narration;
- locally generated symbolic media only when it materially improves the scene.

Do not:

- download copyrighted social-media clips;
- scrape random images;
- fabricate real product screenshots, metrics, or customer results;
- use frames from a nonexistent reference video;
- depend on network media during preview or rendering.

If a specific factual visual cannot be created truthfully without an unavailable asset, replace it with a clear abstract diagram or typography treatment rather than inventing evidence.

## 13. Audio

- Preserve the original dialogue as the authoritative audio.
- Apply conservative noise cleanup only if needed.
- Avoid robotic denoising and aggressive gating.
- Normalize near −14 LUFS integrated with peaks no higher than −1 dBTP, without audible pumping.
- Do not add an arbitrary music track when none is supplied.
- Voice-only is acceptable and preferable to unlicensed music.
- Subtle locally resolved/generated SFX are allowed only for meaningful actions such as a graphic cut, path completion, UI click, or CTA lockup. Keep them well below the dialogue.

## 14. HyperFrames implementation contract

- Delivery canvas: 1080×1920, 9:16.
- Use the source’s measured frame rate when suitable; otherwise use 30000/1001 fps and report the conversion.
- Final delivery: H.264 video + AAC audio MP4.
- Use readable Studio tracks: A-roll picture, dialogue audio, graphic scenes, captions, and optional SFX.
- Use sub-compositions for every major graphic scene.
- Use `class="clip"` and the supported `data-*` timing attributes.
- Use synchronous, paused, registered, seek-safe timelines.
- No autoplay-dependent critical motion.
- No `Date.now()`, `performance.now()`, unseeded `Math.random()`, infinite loops, or timer-driven render-critical animation.
- No render-time network requests.
- Never animate `display`, `visibility`, or `autoAlpha` on a `.clip`; animate a child wrapper.
- Keep IDs unique across the assembled document.
- Maintain social safe zones: essential text approximately within x=80–1000 and y=140–1650, with the CTA comfortably above bottom platform controls.

## 15. Verification workflow

1. Produce `BRIEF.md`, `source-analysis.md`, verified transcript artifacts, `STORYBOARD.md`, and `timing-map.json`.
2. Build the complete project. Do not stop after planning.
3. Search the HyperFrames catalog before hand-building any named effect or transition.
4. Run `hyperframes lint` after the first structural pass and after major timing changes.
5. Run the final `hyperframes check` gate and resolve every lint, runtime, layout, font, motion, and contrast error.
6. Run keyframe diagnostics for the opening push, every punch-in/reframe, and any visual handoff.
7. Inspect snapshots at:
   - exact first frame;
   - 25%, 50%, 75%, and 100% of the opening push-in;
   - every hard-cut boundary ±1 frame;
   - midpoint and completed state of every full-screen interlude;
   - every list/module state;
   - CTA final hold;
   - final-minus-hold and exact final frame.
8. Review the multi-scene animation map.
9. Verify there are no black flashes, missing-font frames, broken Arabic shaping, clipped captions, sub-composition mount flashes, lingering B-roll, reset tails, or output/audio desynchronization.
10. Open the final HyperFrames Studio preview and provide the preview URL for approval. This is the only intended pause.
11. After approval, render delivery quality to `final_reel.mp4`.
12. Verify with `ffprobe`: 1080×1920, expected fps, H.264, AAC, nonzero size, and duration matching the root composition.

## 16. Final delivery report

Deliver:

- the complete HyperFrames project folder;
- `final_reel.mp4` after preview approval;
- `captions.srt`;
- `transcript.json`;
- `timing-map.json`;
- a concise report listing:
  - original and final duration;
  - removed source ranges;
  - number of semantic scenes;
  - actual A-roll/graphic ratio;
  - exact fonts/weights used;
  - registry/media dependencies used;
  - validation status.

## 17. Failure conditions you must not hide

Stop and report a concrete blocker instead of pretending success if:

- no `rawreel.*` file exists or more than one ambiguous candidate exists;
- the video/audio cannot be decoded;
- Arabic transcription remains materially uncertain after review;
- no installed system font renders Arabic legibly with correct shaping;
- Arabic shaping or the selected font loading is incorrect;
- HyperFrames skills/CLI cannot run;
- a required factual visual cannot be created without inventing evidence;
- `hyperframes check` still reports an unresolved error.

The quality bar is a premium editorial reel, not a subtitle template. The result must feel intentionally directed: immediate talking-head hook, fast opening push, small word-timed captions, decisive full-screen editorial interludes, warm tactile paper, symbolic monochrome visuals, high-contrast Arabic hierarchy, restrained warm/cool accents, hard cuts, clean audio, and a confident final CTA.