ugc-unboxing

Trigger

FIRES on a generic unboxing brief: "make an unboxing of ", "open-the-box video", "first look at ", "haul video", "what's inside ". Any subject — the skill is subject-agnostic.

DO NOT FIRE when:

• The user points at one specific unboxing video to reproduce (@template:<slug>, "remix this one", names a slug) → that is the remix path (ralphy template use <slug>), not this skill. See docs/skills-vs-templates.md.
• The request is a different niche (talking-head, tier-list, before/after) → match that niche's skill.

What this skill is

A generalized overlay, not a finished video. It does not name a product, a script, or a cast — it tells the pipeline HOW unboxing videos work so the scenarist and art-director produce a strong one for whatever the user is unboxing. It runs through the normal pipeline; it does not replace intake gates, the reference-required gate, or the quality gates.

Hard invariants

• This skill never bypasses AGENTS.md invariants. The reference-required gate still fires for a named real brand product (an "iPhone 16", a "Dyson Airwrap") — refuse without a ref or logged --no-ref-consent. A no-name / generic product proceeds without a ref.
• All generation routes through ralphy generate (no raw API). Renders go through ralphy render <id>.
• Read MODELS.md before naming any model id — the stack below is a default, not a hardcode.

The niche, in one paragraph

Unboxing lives or dies on the reveal moment and on tactile credibility. The viewer must believe a real human hand is touching a real object. Hands-only or over-shoulder framing (face optional), macro detail on textures/seams, and crisp ASMR-register SFX (cardboard, tape peel, plastic crinkle, the product's own click/snap) do more for retention than any VO line. Hook in the first ~1.5s with the still-sealed box and a tension line ("I cannot believe what's inside this").

Beat structure (default ~15-25s, 9:16)

Beat	Role	~Duration	What it does
1. Sealed-box hook	hook	1.5-3s	Box still closed, hands entering frame, tension line. Scroll-stopper.
2. The open	body	2-4s	The peel / cut / lid-lift. SFX carries this beat — VO optional.
3. Reveal	body	2-4s	First full look at the product, lifted toward camera. The payoff.
4. Detail macro	body	3-6s	Close textures, materials, the one feature worth buying for.
5. Reaction / verdict	body	2-4s	Genuine reaction beat; the one honest opinion line.
6. CTA	cta	1.5-3s	Where to get it / "should I keep it?" engagement bait.

Scale beats up/down with the user's duration. Keep the hook ≤3s regardless.

Framing + lens vocabulary

• Default framing: hands-only, top-down or over-shoulder onto a clean surface. Face is optional — many top unboxing creators never show one.
• Lens: 35-50mm look for the wide; push to a macro register (85-100mm, shallow DOF) for the detail beat.
• Lighting: soft key + fill, slight specular highlight on the product to read material. Avoid flat phone-flash look.
• Surface: one consistent surface across beats — generate a surface/location master plate first (per intake step 3) so every beat matches.
• Anti-AI-slop: real-camera credibility on hands (skin texture, slight asymmetry, natural nail/knuckle detail). See MEMORY.md anti-ai-slop notes; fold the photoreal-hands guideline in if available (ralphy guideline list).

Audio register

• SFX is the star. Tape peel, cardboard, plastic crinkle, the product's own click/snap. Crisp, close, ASMR-leaning.
• VO is sparse — 1-2 honest lines max (hook + verdict). Don't narrate the open; let SFX carry it.
• Music: low instrumental bed under the SFX, or none. Per AGENTS invariant, Kling auto-music is banned in-prompt; music is a separate ElevenLabs Music post-mix in the editor stage.

Default model stack (verify against MODELS.md)

• Keyframes: google/gemini-3-pro-image-preview (nano-banana-pro) for the product + hands anchors.
• i2v: kwaivgi/kling-v3.0-pro — the hand-on-object motion and tactile micro-gestures are its strength. Each beat needs a distinct physical action (peel, lift, rotate) — start/end frames must show a real motion delta (see MEMORY.md start-end-frame note), or clips read static.
• VO: Kling --audio for EN; ElevenLabs for non-EN (confirm target language at intake).
• Captions: per-slot on the locked VO.

Workflow

1. Intake. Run the normal intake (target language, aspect, duration, hard "no"s). Announce: "This is an unboxing — using the unboxing skill." Aspect defaults 9:16.
2. Reference gate. If the product is a named real brand item, the reference-required gate fires — get a ref photo or a logged --no-ref-consent. No-name product → proceed.
3. Scenario. Hand the beat structure above to the scenarist; produce STORYBOARD.md and get the user's "go".
4. Surface master plate first, then product anchor, then per-beat anchors — one beat at a time with checkpoints (intake step 3 cadence).
5. i2v per beat, then VO + SFX, then captions, then ralphy render <id>.
6. Hand off to /evaluator for the post-render quality gate.

Cookbook

• "Make an unboxing of my new coffee grinder" → no-name product, no ref needed. 9:16, ~20s, hands-only, SFX-forward. Match this skill, run the pipeline.
• "First look at the iPhone 16" → named real product → reference gate fires; ask for a ref photo before any generation.
• "Remix the gadget-unboxing-asmr video but use my product" → NOT this skill; this is the remix path → ralphy template use gadget-unboxing-asmr.

See also

• docs/skills-vs-templates.md — why this is a skill and not a template.
• .agents/skills/intake/SKILL.md — the gates this skill runs through.
• MEMORY.md — anti-ai-slop, start↔end frame motion delta, Kling no-music post-mix.