devtool-launch-video

Dev-tool launch video

This template produces 33.8 s at 30 fps, 1920×1080, in 11 scenes. It is built for viewers who have never heard of the product: it shows the pain first, then the product catching exactly that pain, then how to get it. The whole film is driven by a JSON brief. Example: examples/prismor/brief.json (Prismor, an AI-agent runtime security tool). Storyboard: references/storyboard.md.

Workflow

1. Understand the product. Read its README or site. Pull out:

   • the 4–9 concrete dangerous or painful actions it prevents (as real-looking terminal lines)
   • the verdicts it produces
   • 8 features, each one word or phrase plus a line of detail
   • the install commands
   • the tagline, the brand colours and the logo motif

   Use the product's real vocabulary: its rule names, CLI verbs, version.

2. Script. Write the problem lines (rogue_lines) so they escalate, and make the last one the scariest. Make intercept replay the same actions with a verdict each, ending on one ALLOWED so it doesn't read as "blocks everything".

3. Brief. Copy the example brief and replace copy, theme, verdicts, holo (brand light palette) and the assets prompts. Validate with python3 scripts/launch.py check --brief brief.json.

4. Images. Put the brand's real logo at assets/logo.png (square PNG with alpha); never generate or reinterpret it. Then python3 scripts/launch.py assets --brief brief.json generates threat (medium) and holo (low) with gpt-image-2.

5. Check and review. check samples the whole cut and fails on any text overflow. It also warns when copy had to shrink below 80%; shorten that copy. Then look at still --brief … 3 7.5 10.5 14 20 25 29 33 or sheet.

6. Render: render --brief … [--audio track.mp3]. It takes about 4 min on 2 cores. Cuts land on a 0.55 s beat in the features scene, so a ~110 BPM track fits.

Image slots

Slot	Scene	Framing
threat	rogue terminal background (shown at 42% brightness)	dark, ominous, one red light source; the subject on the right third (the terminal covers the left)
logo	reveal, end card, dashboard	the brand's real mark, supplied by the user or the brand site; square PNG with alpha
holo	brand light: split reveal, feature beats, CTA text fill, logo glow	full-frame abstract texture in the brand palette, no objects

Brief keys (copy)

• hook [2]: typed cold-open lines.
• names [n] + names_end: machine-gun list (integrations, platforms).
• rogue_title, rogue_lines [{text, kind}]. kind ∈ agent | inject (red highlight) | cmd ($ …) | leak | dim | ok.
• rogue_captions [[line1, line2] × 3]: line 2 is shown in the danger colour.
• meet, brand, product_sub.
• intercept_title, intercept_caption, intercept [{cmd, verdict, rule, out?}]. out is an extra output line coloured by its verdict. Verdict colours come from top-level verdicts.
• features [8 × {word, sub}]: word ≤ 16 chars.
• dashboard {app, title, nav[6], cards[4 × [label, value]], chart_title, dist_title, table_title, rows[4 × [call, agent, rule, verdict]]} and dashboard_caption.
• install_title, install [{text, kind}], install_agents [names, typed as one line], install_caption.
• tagline, cta, links, license.

Timing: "scenes": {"names": 0} drops a scene; "speed": 1.2 shortens everything.

Design rules

• Never use pills, chips or badges. Verdicts, statuses and CTAs are typeset text (colour, weight, leader lines, rules).
• Every string has a width budget. Long terminal lines word-wrap inside their window, and check must pass before rendering.
• Use the brand's real logo only, and no literal visual puns on the brand name.

Guardrails

• Only use a brand the user owns or has asked for. Keep claims factual: every feature and number shown should come from the product's own docs.
• Fake secrets in terminal lines must be obviously fake. Never paste a real key into a brief.
• Change motion in scripts/launch.py, never in scripts/framesmith/, which is vendored from the repo's core/.