akachandan1/video-studio

Apple / OpenAI-style product launch film or premium spec ad (30–60 s) — minimal supers, big product moments, signature transition, piano/pulse score, end-card lockup. Shots come from code (HyperFrames scenes), Higgsfield AI image→video, or real product footage, all assembled and rendered in HyperFrames. Includes the fold-cut builder (shots hinge shut/open on the centre line).

What is video-studio?

video-studio is a Codex agent skill that apple / OpenAI-style product launch film or premium spec ad (30–60 s) — minimal supers, big product moments, signature transition, piano/pulse score, end-card lockup. Shots come from code (HyperFrames scenes), Higgsfield AI image→video, or real product footage, all assembled and rendered in HyperFrames. Includes the fold-cut builder (shots hinge shut/open on the centre line).

Works with~Claude Code✓Codex CLI~Cursor

npx skills add https://github.com/akachandan1/video-studio/tree/HEAD/.claude/skills/launch-film

Documentation

Launch film (HyperFrames)

1. Concept (get a yes before spending anything)

One idea, one verb. ("Open to more", "Say it.") The product's defining property becomes the film's signature transition or motif (a foldable → fold cuts; a voice product → type that speaks).
Shot plan (table): time, shot, source (code / AI / footage), super text, sound cue. 8–12 shots for 40 s.
Register: big type, few words per super (≤ 6), lots of black, one accent colour, Inter/Instrument Serif (SF Pro is not redistributable). End card: product name, tagline, availability line.
If it's a real brand, label it a spec ad and use only official imagery the user supplies.

2. Shots

Code shots (UI, type, product diagrams): HyperFrames compositions. Best quality-per-cost; zero credits.
AI shots (Higgsfield connector): image first (gpt_image_2_5 ≈ 2.75 credits, nano_banana_pro ≈ 2 credits for accuracy), then image→video (kling3_0 pro, sound off ≈ 7.5 credits / 5 s; seedance_2_5 for people). Use the official product image as reference. Quote the total and wait for a yes. Generate 1 test shot before the batch. Review every clip at full res — AI clips go wrong in ways thumbnails hide (warped product, extra hinges, text gibberish). Trim around artefacts with data-media-start/playback-rate before paying to regenerate.
Footage / renders from the user: conform to H.264, same fps (30).

3. Fold-cut example (`examples/fold-cut/`)

build.py turns a SHOTS list into split-half <video> pairs so any cut can be a fold (outgoing halves hinge shut on the centre line, landing on a magnetic "clack") or an unfold (incoming halves open), plus SUPERS (hero / lower supers with line reveals). Copy it into videos/<name>/, put clips in assets/clips/, edit SHOTS and SUPERS, python3 build.py, lint, check, render. The same build-from-a-list pattern works for any signature transition (iris, slide, light-leak).

4. Sound

score skill → examples/launch_piano_120bpm_40s.py: felt piano motif over a warm pulse with a clack on every fold. Change the cue list to your cuts. Or the Fal music-video plugin / a licensed track + beats.py.

5. Finish

npx hyperframes check (contrast, overlaps), snapshot every super, render, loudnorm −14, contact-sheet 6 frames and look before sending. Offer 9:16 cut-down (re-layout supers; don't just crop).