| name | video-factory-concept |
| description | Brainstorm distinct video concepts for a brief — cinematic videos, brand films, motion accents, and background loops for product websites. Produces 3–5 genuinely different creative directions (not five flavors of one safe idea), each with a premise, the feeling it creates, why it fits the product and placement, and how hard it is to pull off in Higgsfield. Use this skill any time the work starts with "what could this video be" — concepting a hero background loop, a launch film, a section divider, an explainer, or a social spot — even when the user just says "I need a video for X." First stage of the Video Factory chain; hands the chosen concept to video-factory-storyboard. Pulls toward concrete subjects over the drone-over-city / glowing-AI clichés that AI video defaults to.
|
| type | skill |
| project | skills-library |
| plugin | video-factory |
| aliases | ["video-concept","video-ideation","concept-brainstorm"] |
| tags | ["type/skill","plugin/video-factory","scope/specialty","topic/video","topic/ideation"] |
| status | active |
Video Factory — Concept
The job here is divergence. Most AI-generated video concepting collapses to one idea ("an abstract animation representing data flowing") dressed five ways. A useful concept set gives the user genuinely different bets — different subjects, different emotional registers, different production approaches — so the choice that follows is real.
Load ../../references/video-factory-foundations.md for output types, registers, and the motion anti-clichés. Don't re-derive them here.
When to invoke
- The brief is set (or roughly known) and the user wants directions to choose from.
- Someone says "I need a video for [product/page]" and hasn't decided what it is yet.
- An existing video isn't landing and needs fresh angles.
If there's no brief at all, get the four things that change everything first (below), then concept. Don't make the user fill out a form — ask only what you can't infer.
The four things a concept needs to respect
You can infer most of a brief, but these four genuinely change the concept and are worth confirming if unknown:
- Placement — where does it play? A silent hero loop behind a headline is a different object from a sound-on launch film. Placement sets the constraints (see the output-type table in foundations).
- Product truth — what does the thing actually do, and for whom? Concepts anchored in the real product beat concepts anchored in a metaphor.
- Feeling — what should a viewer feel in the first two seconds? Name it specifically ("quiet competence," "kinetic momentum," not "professional").
- Brand — Slalom, Composable DXP, or client custom. Governs palette and motion language.
What good concepts look like
Produce 3–5 concepts that disagree with each other. Force range deliberately:
- One should be the straight, safe read — the concept a competent team would ship. It's the baseline, and sometimes it's right.
- One should be subject-led — a concrete, real thing in frame (a hand, a material, a place, the product in use) rather than an abstraction.
- One should be abstract / motion-led — texture, light, material, or graphic motion, no literal subject. (This is often the strongest answer for a silent background loop where readability matters.)
- One should be a stretch — the idea that makes the room lean in, higher risk, maybe harder to generate. Even if it's not chosen, it pulls the others up.
- Optionally a format-twist — same message, unexpected container (a single unbroken take, a loop that hides its seam, a split-screen, a match-cut series).
Each concept must be honest about Higgsfield feasibility. The models are image-to-video and strongest at atmospheric, single-subject, camera-driven shots; weakest at precise text, specific brand logos, real named-person likeness, and complex multi-actor choreography. Flag where a concept fights the tool.
Output format
Write 10_Concepts.md in the run's WIP folder. For each concept:
## Concept N — [evocative name]
**Premise.** One or two sentences: what the viewer sees, start to finish.
**Feeling.** The two-second emotional read. Specific.
**Why it fits.** The line connecting this to the product truth and the placement.
**Register.** Which motion register from foundations (editorial-tech, premium-restrained, kinetic, material, anamorphic-cinematic).
**Higgsfield feasibility.** Easy / Moderate / Hard, plus the one thing that's tricky (looping seam, person likeness, rendered text, multi-shot) and how to handle it.
**Reference touchstone.** A film, ad, or visual style it rhymes with, so the user can picture it.
Close with a one-line recommendation — which you'd ship and why — and an invitation to pick or remix. The user's preference is an opinionated default plus the fork, not a neutral menu.
Examples
Example 1 — concept range for a silent hero loop
Brief: background video for a composable-commerce platform's homepage hero; headline overlays center; audience is digital-experience leaders; feeling = "calm control of complexity"; Slalom Composable DXP brand.
- Concept A — "Assembled" (subject-led, Moderate). Modular glass panels drift together into a single clean surface, slow, in shallow focus. Literalizes "composable" without a diagram. Tricky: keeping motion low enough behind the H1 — keep the assembly in the lower third, headline safe-area top-center.
- Concept B — "Current" (abstract/material, Easy). Macro of a slow-moving liquid-metal surface catching cool light, endless and seamless. Pure texture, reads as premium, loops cleanly with matched start/end frame. The strongest background bet.
- Concept C — "One Take" (stretch, Hard). A single continuous push through a series of architectural thresholds that each reframe to a new brand color. Striking, but multi-shot continuity is where the models strain — would need careful end-frame chaining.
Example 2 — three bets for a launch film
Brief: 20-second launch film for a developer tool, plays with sound on a landing page; feeling = "fast, in control."
- Subject-led: over-the-shoulder of hands at a mechanical keyboard, screen reflected in glasses, cut to the result shipping — concrete, human, momentum. (Watch likeness/text on screen.)
- Abstract: command-line text resolving into a flowing build graph rendered as light, kinetic register. (Rendered text → route keyframe to chatgpt image 2.)
- Format-twist: the whole film is one unbroken match-cut loop so it can also run muted as a background — two deliverables from one idea.
Constraints
- Never hand back one idea with cosmetic variants. If the concepts don't disagree, they're not done.
- Anchor at least one concept in a concrete subject. Abstraction is a choice, not a default.
- Be honest about feasibility — a beautiful concept the tool can't render wastes the user's credits and trust.
- For a silent background loop, weight the set toward low-motion, readable, loop-friendly ideas. Spectacle that drowns the headline is the wrong answer even if it's the prettiest.
- Hand off cleanly: the chosen concept's name, register, and feasibility notes flow into
video-factory-storyboard.
