System Requirements Document for hyper-hi

1. Introduction

hyper-hi is an AI video-editing agent built for a single working content creator. The product's intent is direct and personal: the creator hands their own video footage to an AI agent and receives edited video back, without needing to operate a conventional timeline-based editor themselves. The agent is the collaborator; the creator is the director who supplies footage, states what they want, and receives a finished cut.

The audience is the requesting creator — Naveed R.B, a content creator in Pakistan who produces their own videos and wants the repetitive, time-consuming part of editing handled by AI assistance. The product is a creative instrument for one person's ongoing production workflow, not a multi-tenant studio tool, not a team collaboration platform, and not a marketplace.

The product must feel like a creative tool with attitude: energetic, deadline-driven, poster-loud, closer to a late-night edit bay than to a calm enterprise dashboard.

2. System Overview

hyper-hi is a first-party web application with application-owned identity and custom UI. The creator signs up once, uploads source video, directs the AI agent, and receives edited results that persist across sessions so they can return, review, and continue.

Current delivery. A custom web application with six destinations: an anonymous Landing page that explains the product and introduces the editing workflow, Sign Up and Login for first-use enrollment and returning verification, and three protected destinations — Projects, Upload, and Results — where the creator's ongoing editing work, source video intake, and AI-edited outputs live.

Actors.

• Content Creator (human, active) — the sole accepted persona. Initiates all editing work, supplies footage, directs the agent, and receives results.
• AI editing agent (system actor, non-persona) — performs the video editing work on the creator's behalf and produces the edited result. It is not a human participant and has no independent goals.
• Application backend (system actor, non-persona) — stores projects, source videos, and edited results; enforces access to protected destinations.

Accepted behavior. The creator can enroll, verify on return, upload source video, direct the AI agent to edit it, and receive and continue with an edited result — repeatedly, as part of an ongoing content production workflow.

Narrow exclusions. No team or multi-user collaboration, no client/agency review workflow, no publishing or distribution to external platforms, no monetization or billing surface, no social or community features, and no third-party editing-suite integration are accepted. Nothing in this document should be read as authorizing those.

2a. Product Interpretation and Delivery Boundary

Delivery ownership. hyper-hi is delivered as a first-party custom web application. All six destinations are application-owned and application-rendered. There is no provider-owned surface, no external destination, and no headless-only delivery in the accepted scope.

Access ownership. Identity is application-owned. The creator's projects, source videos, and edited results are durable, creator-specific state that must remain bound to the correct person and be resumable across sessions. This makes identity continuity indispensable rather than optional: without it, a returning creator could not reach their own stored work.

Consequently:

• Landing is anonymously reachable. It is the public entry surface and explains the product before any identity exists.
• Sign Up and Login are anonymously reachable. A protected destination cannot own the interaction that establishes access to itself, so first-use enrollment and returning verification are their own access boundary.
• Projects, Upload, and Results require an established identity. Their contents are creator-specific and cannot be shown to an anonymous visitor.

Current vs. future. Everything described in Sections 3 through 5 is current. No future-horizon requirements were accepted in the authoritative thread; Section 11 records the boundaries that keep the current scope honest.

2b. Source Content Inventory

Not applicable. No reference directive in this project declares content_source authority, so no source content inventory is produced.

2c. Page Content and Component Coverage

Landing

• Information and state. Anonymous public entry. Explains what hyper-hi is: an AI agent that edits a content creator's videos. Introduces the editing workflow — upload footage, direct the agent, receive an edited result. No creator-specific data is shown, because no identity exists yet.
• Primary action. Begin the editing workflow, which routes an unenrolled visitor to Sign Up and a returning creator to Login.
• Supporting actions. Navigate to Login directly; read the workflow explanation; move between landing sections.
• Domain entities. None owned. The page describes the product and the workflow; it holds no project, video, or result state.
• Component responsibilities.
  • Full-bleed poster hero: stacked all-caps headline set as the hero image itself, with a red block cutting across the middle line and a yellow marquee band beneath carrying process words.
  • Primary call-to-action: sharp-cornered red rectangle with a yellow offset shadow, pinned bottom-left, reading as the start of the editing workflow.
  • Workflow explanation section: the three-step shape of the product (provide footage → direct the agent → receive the edit), presented as colliding type and colour blocks.
  • Marquee band reused as a section divider between landing sections.
  • Fixed left navigation rail of stacked uppercase labels; becomes a horizontal marquee bar on mobile.
• States.
  • Loading: static content; no data fetch is required, so no loading state is presented.
  • Empty: not applicable — the page has no collection.
  • Success: the visitor understands the product and reaches either Sign Up or Login.
  • Error: if a navigation target fails to resolve, the visitor remains on Landing with the call-to-action still available to retry.
  • Recovery: the call-to-action and the direct Login link remain reachable at all times, so a failed navigation never strands the visitor.

Login

• Information and state. Anonymous access surface for a returning creator. Collects the credentials that verify an existing identity so the creator can reach their stored projects, source videos, and edited results.
• Primary action. Verify identity and continue into the protected workspace.
• Supporting actions. Move to Sign Up if the visitor has no identity yet; return to Landing.
• Domain entities. Creator identity (verification only — no project, video, or result data is displayed here).
• Component responsibilities.
  • Credential entry form with clearly labelled fields.
  • Submit control styled as a sharp-cornered red rectangle with a yellow offset shadow.
  • Inline validation messaging for missing or malformed input.
  • Link across to Sign Up.
• States.
  • Loading: submit control shows an in-progress state while verification is attempted; the form is not re-submittable during this window.
  • Empty: the form renders with empty fields and no error messaging.
  • Success: verification succeeds and the creator is taken to Projects.
  • Error: verification fails; a clear message is shown and the entered identity value is preserved so the creator can correct it without retyping everything.
  • Recovery: the creator can retry immediately, or switch to Sign Up if they do not yet have an identity.

Sign Up

• Information and state. Anonymous access surface for a first-time creator. Establishes the identity that will own the creator's editing workspace and durable video state.
• Primary action. Create the identity and enter the workspace.
• Supporting actions. Move to Login if an identity already exists; return to Landing.
• Domain entities. Creator identity (creation only).
• Component responsibilities.
  • Enrollment form with clearly labelled fields and stated input expectations.
  • Submit control styled as a sharp-cornered red rectangle with a yellow offset shadow.
  • Inline validation messaging for missing, malformed, or already-taken input.
  • Link across to Login.
• States.
  • Loading: submit control shows an in-progress state while the identity is created; the form is not re-submittable during this window.
  • Empty: the form renders with empty fields and no error messaging.
  • Success: the identity is created and the creator is taken to Projects, which is empty until they upload their first video.
  • Error: creation fails; a specific message identifies what needs correcting and the entered values are preserved.
  • Recovery: the creator can correct and resubmit, or switch to Login if an identity turns out to already exist.

Projects

• Information and state. Protected, creator-specific overview of ongoing video-editing work and previously produced results. Shows the creator's own projects only. Each project entry carries its identity, its current state in the editing lifecycle, and its associated source video and edited result where those exist.
• Primary action. Open a project to continue with it — either to supply footage or to reach its edited result.
• Supporting actions. Start a new project, which leads into Upload; return to a completed project's edited result in Results; scan the project list.
• Domain entities. Project (identity, state, timestamps, associated source video, associated edited result).
• Component responsibilities.
  • Dense poster grid of sharp rectangular project tiles, each with an oversized index number and a hard-cropped 16:9 thumbnail carrying a red overlay bar with the project title.
  • Per-tile state indicator showing where the project sits in the editing lifecycle.
  • New-project control leading to Upload.
  • Marquee header band above the grid.
  • Fixed left navigation rail; horizontal marquee bar on mobile.
• States.
  • Loading: grid placeholders hold the poster layout while the creator's projects are retrieved.
  • Empty: a first-time creator sees an explicit empty state that names the next action — upload a video to start the first project — with the new-project control as the single obvious path.
  • Success: the creator's projects render as tiles with their current state, and each tile opens.
  • Error: retrieval fails; a clear message is shown with a retry control, and the new-project path remains available so the creator is not blocked from starting work.
  • Recovery: retry reloads the grid; if retrieval continues to fail, the creator can still begin a new project through Upload.

Upload

• Information and state. Protected, focused destination where the creator provides video content to the AI agent. Holds the selected source video, its upload progress, and the creator's direction for how the agent should edit it.
• Primary action. Submit the source video and the editing direction to the AI agent, which starts the edit.
• Supporting actions. Select or replace the source video; enter or revise the editing direction; cancel before submitting; return to Projects.
• Domain entities. Source video (file, size, duration, upload state); Project (the container the upload belongs to); Editing direction (the creator's stated instruction to the agent).
• Component responsibilities.
  • Video selection control accepting the creator's own footage, with the selected file's name and size shown back.
  • Upload progress indicator with a truthful in-progress state.
  • Editing-direction input where the creator states what they want done to the video.
  • Submit control styled as a sharp-cornered red rectangle with a yellow offset shadow.
  • Cancel control that abandons the in-progress upload without creating a project.
• States.
  • Loading: upload progress is shown while the source video transfers; the submit control is unavailable until the transfer completes.
  • Empty: no video selected — the selection control is the only active element, with the editing-direction input and submit control unavailable.
  • Success: the source video and editing direction are accepted, the project is created or updated, and the creator is taken to Results to follow the edit.
  • Error: the transfer fails, the file is rejected, or the submission is refused; a specific message states what went wrong, the selected file and any entered direction are preserved, and the creator can retry without re-entering their work.
  • Recovery: retry the transfer or submission; replace the file; or cancel out to Projects with nothing created.

Results

• Information and state. Protected, focused destination where the creator receives and continues with an AI-edited video result. Shows the edited output for a project, the project it belongs to, and the state of the agent's work.
• Primary action. Take the edited result forward — review it and continue with it as the outcome of that project.
• Supporting actions. Return to the project in Projects; start another edit through Upload; revisit a previously produced result.
• Domain entities. Edited result (the agent's output video, its state, its association to the project and source video); Project.
• Component responsibilities.
  • Edited-result presentation area with a hard-cropped 16:9 frame and a red overlay bar carrying the project title.
  • Agent-work state indicator showing whether the edit is in progress, complete, or failed.
  • Project context showing which source video and direction produced this result.
  • Continuation controls: back to the project, and start another edit.
• States.
  • Loading: the agent's work is in progress; the result area shows an explicit in-progress state rather than an empty frame.
  • Empty: no result exists yet for the project; the page states that the edit has not produced output and offers the path back to the project.
  • Success: the edited result is available and presented, with its project context, and the creator can continue with it.
  • Error: the agent's edit fails; a clear failure message is shown with the project context intact and a path to retry the edit through Upload.
  • Recovery: retry the edit, return to the project, or start a new edit — the creator is never left on a dead end.

3. Functional Requirements

FR-1 — Enroll for first use

As a Content Creator I should be able to establish my own identity on first use so that my editing workspace and video state belong to me and persist.

• Provenance: required_inference
• Trigger / input: the creator arrives at Sign Up from Landing and submits enrollment details.
• Observable result: an identity is created and the creator lands in Projects, which is empty until their first upload.
• Access state: anonymous entry; the interaction that establishes access is not itself protected.
• Failure / recovery: if enrollment is refused, a specific message identifies what to correct, entered values are preserved, and the creator can resubmit or switch to Login.
• Continuation: the creator proceeds to Upload to provide their first video.

FR-2 — Verify identity on return

As a Content Creator I should be able to verify my identity when I come back so that I can reach my stored projects, source videos, and edited results.

• Provenance: required_inference
• Trigger / input: the creator arrives at Login and submits their credentials.
• Observable result: verification succeeds and the creator reaches Projects with their own work intact.
• Access state: anonymous entry; protected state remains unavailable until verification succeeds.
• Failure / recovery: if verification fails, a clear message is shown, the entered identity value is preserved, and the creator can retry or switch to Sign Up.
• Continuation: the creator resumes their ongoing editing work from Projects.

FR-3 — Provide source video to the AI agent

As a Content Creator I should be able to upload my own video footage so that the AI agent has the material it needs to edit.

• Provenance: explicit
• Trigger / input: the creator selects a video file on Upload.
• Observable result: the selected file is shown back with its name and size, and its transfer progress is visible.
• Access state: requires an established identity.
• Failure / recovery: if the transfer fails or the file is rejected, a specific message states why, the selection is preserved, and the creator can retry or replace the file.
• Continuation: once the transfer completes, the creator states their editing direction and submits.

FR-4 — Direct the AI agent's edit

As a Content Creator I should be able to state how I want my video edited so that the AI agent's output matches what I actually want.

• Provenance: explicit
• Trigger / input: the creator enters their editing direction on Upload and submits it with the source video.
• Observable result: the source video and the direction are accepted together, and the project is created or updated with both.
• Access state: requires an established identity.
• Failure / recovery: if submission is refused, a specific message is shown, the selected file and entered direction are both preserved, and the creator can retry without re-entering their work.
• Continuation: the creator is taken to Results to follow the agent's work.

FR-5 — Have the AI agent edit the video

As a Content Creator I should have the AI agent perform the edit on my footage so that I get an edited video without doing the cutting myself.

• Provenance: explicit
• Trigger / input: an accepted source video and editing direction for a project.
• Observable result: the agent's work is visible as an in-progress state on Results, and an edited result is produced for that project.
• Access state: requires an established identity; the result belongs to the creator's own project.
• Failure / recovery: if the edit fails, Results shows a clear failure state with the project context intact and a path to retry through Upload.
• Continuation: on success the creator reviews the edited result and continues with it.

FR-6 — Receive and continue with the edited result

As a Content Creator I should be able to receive my AI-edited video and take it forward so that the edit becomes usable output for my content.

• Provenance: explicit
• Trigger / input: the agent completes an edit for one of the creator's projects.
• Observable result: the edited result is presented on Results with its project context — which source video and which direction produced it.
• Access state: requires an established identity; only the creator's own results are shown.
• Failure / recovery: if no result exists yet, the page states that plainly and offers the path back to the project; if the edit failed, the failure is stated with a retry path.
• Continuation: the creator returns to the project in Projects or starts another edit through Upload.

FR-7 — Revisit ongoing and past editing work

As a Content Creator I should be able to see my ongoing and previously produced editing work in one place so that I can pick up where I left off.

• Provenance: required_inference
• Trigger / input: the creator opens Projects after verifying their identity.
• Observable result: their own projects render as tiles, each showing its current state in the editing lifecycle and its associated source video and edited result where those exist.
• Access state: requires an established identity; the list contains only the creator's own projects.
• Failure / recovery: if retrieval fails, a clear message and a retry control are shown, and the new-project path stays available so the creator is not blocked from starting work.
• Continuation: the creator opens a project to continue it, or starts a new one through Upload.

FR-8 — Understand the product before enrolling

As a Content Creator I should be able to understand what hyper-hi does before I sign up so that I know the AI agent will edit my videos the way I need.

• Provenance: required_inference
• Trigger / input: an anonymous visitor arrives at Landing.
• Observable result: the page explains that hyper-hi is an AI agent that edits a content creator's videos, and introduces the workflow of providing footage, directing the agent, and receiving an edit.
• Access state: anonymous; no creator-specific data is shown.
• Failure / recovery: if a navigation target fails to resolve, the visitor stays on Landing with the call-to-action still available to retry.
• Continuation: the visitor proceeds to Sign Up, or to Login if they already have an identity.

4. User Personas

Content Creator

Product context. Naveed R.B is a content creator in Pakistan who produces their own videos and handles the editing themselves. Editing is the bottleneck in their production rhythm: it is the slow, repetitive part that sits between having footage and shipping content. They are not looking for a professional timeline editor to operate — they are looking for an agent that does the cutting so they can keep producing.

Primary goal. Get their own video footage edited by an AI agent, repeatedly, as part of an ongoing content production workflow — without doing the edit by hand.

Distinct accepted responsibilities.

• Enrolling once so their workspace and video state belong to them and survive between sessions.
• Verifying their identity on return so they can reach work they already started.
• Selecting and providing their own source video footage to the agent.
• Stating how they want that footage edited — the direction that shapes the agent's output.
• Reviewing the AI-edited result and deciding what to do with it next.
• Revisiting ongoing and previously produced work to pick up where they left off.

Relevant inputs and decisions. Which video file to hand over; what editing direction to give the agent; whether a produced result is what they wanted or needs another pass; whether to continue an existing project or start a new one.

Interactions with other accepted participants. The creator is the only human participant. They interact with the AI editing agent, which performs the edit on their footage and produces the result they receive, and with the application backend, which stores their projects, source videos, and edited results and keeps them bound to their identity. Neither is a human counterparty; both exist to serve the creator's work.

Observable success. The creator uploads footage, states a direction, and receives an edited video back that they can continue with — and can come back later to find that work still there.

What makes this role distinct. The creator is simultaneously the director and the sole beneficiary of the output. There is no client to satisfy, no reviewer to route through, and no team to coordinate with. Every decision in the product is theirs alone, which is why the product is built as a fast personal instrument rather than a collaborative workspace.

5. Core User Flows

Flow A — First use: enroll and start the first edit

1. The creator arrives at Landing as an anonymous visitor and reads what hyper-hi does: an AI agent that edits a content creator's videos.
2. The creator reads the workflow explanation — provide footage, direct the agent, receive an edit — and chooses the primary call-to-action to begin editing.
3. Because no identity exists yet, the creator is taken to Sign Up.
4. The creator submits enrollment details. The submit control shows an in-progress state while the identity is created.
5. Observable result: the identity is created and the creator lands in Projects, which shows an explicit empty state naming the next action — upload a video to start the first project.
6. The creator follows the new-project control into Upload.
7. The creator selects their own video file. The file's name and size are shown back and its transfer progress is visible.
8. Once the transfer completes, the creator enters their editing direction — what they want the agent to do to the footage — and submits.
9. Observable result: the source video and direction are accepted together, the project is created, and the creator is taken to Results.
10. Failure / recovery: if enrollment is refused at step 4, a specific message identifies what to correct, the entered values are preserved, and the creator can resubmit or switch to Login. If the transfer fails or the file is rejected at step 7, the selection is preserved and the creator can retry or replace the file. If submission is refused at step 8, both the file and the entered direction are preserved so nothing is retyped.
11. Next step: the creator follows the agent's work on Results (Flow C).

Flow B — Returning: verify and resume existing work

1. The creator returns to hyper-hi and goes to Login.
2. The creator submits their credentials. The submit control shows an in-progress state while verification is attempted.
3. Observable result: verification succeeds and the creator reaches Projects, where their own projects render as tiles, each showing its current state in the editing lifecycle and its associated source video and edited result where those exist.
4. The creator scans the grid and opens the project they want to continue.
5. Failure / recovery: if verification fails at step 2, a clear message is shown, the entered identity value is preserved, and the creator can retry immediately or switch to Sign Up if they do not yet have an identity. If the project list fails to load at step 3, a clear message and a retry control are shown, and the new-project path stays available so the creator is not blocked from starting work.
6. Next step: the creator continues the opened project — supplying footage through Upload if it has none, or reaching its edited result in Results.

Flow C — Direct the agent and receive an edited result

1. Starting from Upload with a source video selected and an editing direction entered, the creator submits the work to the AI agent.
2. Observable result: the project is created or updated with both the source video and the direction, and the creator is taken to Results.
3. On Results, the creator sees the agent's work as an explicit in-progress state rather than an empty frame, alongside the project context showing which source video and which direction produced this result.
4. The AI editing agent performs the edit on the creator's footage.
5. Observable result: the edited result becomes available and is presented on Results with its project context intact.
6. Failure / recovery: if the agent's edit fails, Results shows a clear failure state with the project context preserved and a path to retry the edit through Upload. If no result exists yet for the project, the page states that plainly and offers the path back to the project.
7. Next step: the creator continues with the edited result — returning to the project in Projects, or starting another edit through Upload.

Flow D — Revisit and continue previously produced work

1. The creator verifies their identity at Login and reaches Projects.
2. Observable result: their ongoing and previously produced editing work renders as a dense poster grid of tiles, each with its current lifecycle state and its hard-cropped thumbnail carrying the project title.
3. The creator opens a project whose edited result already exists.
4. Observable result: the creator reaches that project's edited result on Results, presented with the source video and direction that produced it.
5. Failure / recovery: if the project list fails to load at step 2, a clear message and a retry control are shown, and the new-project path remains available.
6. Next step: the creator continues with the result, or starts a new edit through Upload.

6. Visuals, Colors and Theme

Muse and headline. Paula Scher. Typography as editing architecture — words are the dominant visual material, flat colour blocks collide, edges are hard, and contrast is poster-scale. Editing is itself a rhythm of cuts, beats and collisions, so the interface is built from the same material.

Mode. Dark mode only. The product lives on a dark ink ground.

Colour tokens (exact).

Proportion. 60% dark ground, 25% red blocks, 10% yellow accents, 5% off-white text. No blue anywhere in the palette.

Typography.

• Headings: Anton, weight 400, all-caps, tight leading 0.9, tracking -0.02em. Headlines run 72px–160px and are set as full-width type blocks, not labels above content.
• Body: Oswald, weight 300/400, 16–20px, line-height 1.6 for readability against the dark ground.
• Scale: 1.5 modular — 160 / 96 / 64 / 40 / 24 / 18 / 16.
• Responsive: mobile display sizes clamp down to 56–72px; desktop headlines reach 128–160px.

Shape language. Hard edges and flat colour blocks. No rounded cards, no soft shadows, no glass. Rectangles collide and overlap; type blocks sit directly on colour fields. Borders are 2px solid ink or off-white when separation is needed. Diagonal bands and offset rectangles serve as section dividers. Buttons are sharp-cornered rectangles with a 2px offset shadow in the accent colour — a printed-poster move, not a UI lift.

Layout. Poster grid with asymmetric columns. The hero is a full-bleed type composition: a stacked headline spanning the viewport, a red block cutting across it, and a yellow accent band underneath. Sections are separated by hard colour wipes, not whitespace. The projects list is a dense poster grid of rectangular tiles, each with an oversized index number and a hard-cropped video thumbnail. Navigation is a fixed left rail of stacked uppercase labels that becomes a horizontal marquee bar on mobile. No centred SaaS hero, no grid of identical hover-lift cards.

Imagery. Typography is the image. Where photography appears it is high-contrast, hard-cropped and treated as a duotone — red/off-white or yellow/ink — so it reads as a printed poster element, not a stock photo. Video thumbnails in the projects grid are cropped to 16:9 with a red overlay bar carrying the project title. No 3D renders, no gradient blobs, no soft illustration.

Readable content stays whole. Headlines, wordmarks, labels, numbers, card text and controls stay entirely inside the viewport and their container at 375px, 768px and 1280px, wrapping or scaling (for example font-size: clamp(...) with its mobile size) to fit, and nothing covers any part of them. Imagery, decoration and motion may be cropped, bled off an edge, rotated, overlapped or cut as the direction asks, provided they cover no readable text or control. Marquees and scrollable rows may cross the viewport edge by design; every item must become fully readable as it passes.

7. Signature Design Concept

The hero is the type. On Landing, a full-viewport poster composition sits on the near-black ground #0E0E10. A single stacked headline in Anton, all-caps, spans the full width — EDIT FASTER / CUT HARDER / SHIP DAILY — each line set at 96–160px and flush left, with the lines at deliberately different sizes so the stack reads as a composition rather than a paragraph. A solid red #E63A2E rectangle cuts diagonally across the middle line, colliding with the words rather than sitting behind them. Beneath the headline, a mustard #F2C94C horizontal band carries a scrolling marquee of creator-facing process words — UPLOAD • PROMPT • RENDER • EXPORT — which repeats later as a section divider between landing sections.

The primary call-to-action is a sharp-cornered red rectangle pinned bottom-left with a 4px yellow offset shadow, reading START EDITING. It is the only control in the composition, and it is unmistakable.

There is no centred headline, no blue button, no gradient blob. The type bleeds off the right edge at desktop scale and clamps down to 56–72px on mobile, where the fixed left rail becomes a top horizontal marquee. The composition recomposes only accepted content — the product's explanation, its three-step workflow, and the path into Sign Up or Login — and introduces no new behaviour.

8. Interaction Model & Motion Direction

Interaction Model: Animated Motion Tempo: expressive Hero Dimensionality: flat

Landing Hero Motion Brief.

• Focal subject. The stacked Anton headline itself — EDIT FASTER / CUT HARDER / SHIP DAILY — with the red block cutting across the middle line and the yellow marquee band beneath it.
• Input → transformation → outcome thesis. As the visitor arrives, each word of the headline drops in from the top with a hard cut, no easing theatrics, stacking into the full composition; the red rectangle then sweeps across the middle line and settles into its diagonal cut; the yellow band beneath begins its horizontal marquee of process words. The outcome is a fully composed poster in which the type is the hero image and the single red call-to-action is the obvious next move.
• Motion vocabulary. Hard cuts, staggered word reveals, horizontal marquees, colour-block wipes. No easing theatrics, no lift, no shadow growth.
• Composed first frame. Before any motion, the composition already reads: dark ground, the headline stack in place at its final sizes, the red block at its diagonal cut, the yellow band present, and the red call-to-action with its yellow offset shadow pinned bottom-left. The first frame is a finished poster, not a blank stage.
• Reduced-motion state. Marquees and colour-block wipes are removed. The hero becomes static stacked type with the red block and yellow band in their final positions, and the projects row becomes a horizontally scrollable row so every item can be brought fully into view. Nothing readable is lost.

Motion across the product.

• Type marquees run horizontally on Landing and in the Projects header.
• Section transitions use colour-block wipes — a red rectangle sweeps across the viewport before the next section appears.
• Hover on a project tile flips the thumbnail to a red duotone and slides the title up from the bottom edge with a hard cut, no lift or shadow growth.
• Navigation is a fixed left rail of stacked uppercase Oswald labels, each with a 2px red underline that wipes in on hover; on mobile the rail becomes a top horizontal marquee.
• Buttons and primary actions use a 4px offset solid shadow in the accent colour instead of a soft drop shadow.

9. Non-Functional Requirements

• NFR-1 — Identity-bound state. Projects, source videos, and edited results are durable and creator-specific, and must remain bound to the correct creator across sessions. Provenance: required_inference — the accepted workflow is ongoing and revisitable, which is impossible without continuity.
• NFR-2 — Protected destinations. Projects, Upload, and Results are reachable only with an established identity. Landing, Sign Up, and Login are anonymously reachable. Provenance: explicit planning access contract.
• NFR-3 — Upload progress truthfulness. A source video transfer must expose a real in-progress state, and the submit control must not be available until the transfer completes. Provenance: required_inference — the creator must not be able to submit work the agent has not received.
• NFR-4 — Failure preservation. On any failure in enrollment, verification, upload, submission, or editing, the creator's entered values and selected file are preserved so no work is retyped. Provenance: required_inference — recovery must not cost the creator their input.
• NFR-5 — Readable content integrity. Headlines, labels, numbers, card text and controls remain whole and uncovered at 375px, 768px and 1280px. Provenance: explicit creative direction constraint.
• NFR-6 — Reduced motion. With prefers-reduced-motion, marquees and wipes are removed and a usable static arrangement is provided, with scrollable rows allowing every item to be brought fully into view. Provenance: explicit creative direction constraint.
• NFR-7 — Palette discipline. No blue or indigo appears anywhere in the product. Provenance: explicit creative direction constraint.
• NFR-8 — Backend integration. The product requires backend integration to store projects, source videos, and edited results and to enforce access to protected destinations. Provenance: explicit planning contract.

10. Tech Stack

• Frontend: React — a custom web UI is required, and the poster-grid layout, marquees, and colour-block wipes are all custom-rendered.
• Backend: Python / FastAPI — serves the creator's projects, source videos, edited results, and identity verification.
• Storage: persistent storage for creator identities, project records, uploaded source videos, and produced edited results. Video files are stored as binary objects; project and result records are stored relationally.
• AI editing agent: a backend-integrated video-editing service that accepts a source video and an editing direction and produces an edited result. Its internal implementation is not constrained by this document.
• Containerization: Docker / docker-compose for local and single-host deployment.
• Orchestration: Kubernetes is not required by any accepted requirement and is not included.

11. Assumptions and Constraints

Assumptions.

• A-1. The creator is the sole human user of the product. No team, client, reviewer, or collaborator role is accepted. (Assumption — narrow, consistent with the single-persona contract.)
• A-2. The AI editing agent is a system actor, not a human participant. It has no independent goals and produces output only in response to the creator's submitted source video and direction. (Assumption — narrow.)
• A-3. The creator supplies their own footage; the product does not source, license, or generate video content on the creator's behalf. (Assumption — narrow.)
• A-4. The creator's editing direction is expressed as text input alongside the uploaded video. (Assumption — the source states the creator directs the agent but does not name the input mechanism; text is the narrowest sufficient form.)

Constraints.

• C-1. Identity is application-owned. Self-service enrollment and returning verification are the only accepted access paths; no invitation, provisioning, or deployment bootstrap is accepted. (Explicit planning contract.)
• C-2. The six destinations — Landing, Login, Sign Up, Projects, Upload, Results — are the complete and final page inventory. No page may be added, removed, merged, split, renamed, reordered, or have its access changed. (Explicit versioned page contract.)
• C-3. The Content Creator is the complete and final persona catalog. No persona may be added, removed, merged, renamed, or replaced. (Explicit persona contract.)
• C-4. No blue or indigo anywhere; no white or near-white page backgrounds; no rounded cards, pill buttons, glassmorphism, gradient blobs, or soft drop shadows; no Inter, Roboto, Arial, Helvetica, Open Sans, Lato, Poppins, or system-ui for headings or body; no centred SaaS hero; no grid of identical hover-lift cards; no 3D renders, particle fields, or WebGL scenes; no decorative illustration or friendly character art. (Explicit creative direction.)
• C-5. No team collaboration, client/agency review workflow, publishing or distribution to external platforms, monetization or billing surface, social or community features, or third-party editing-suite integration is accepted. (Narrow exclusions derived from the accepted scope.)
• C-6. No future-horizon requirements were accepted. Anything not described in Sections 3 through 5 is out of current scope. (Boundary statement.)

Presentation and technology defaults.

• [Default — not specified by user] The specific AI video-editing model or service behind the agent is unspecified; any implementation satisfying FR-5 is acceptable.
• [Default — not specified by user] Accepted source video formats and maximum file size are unspecified; the implementation should accept common creator-facing video formats and state its limits in the Upload interface.
• [Default — not specified by user] Credential requirements for enrollment are unspecified; the implementation should state its input expectations inline on Sign Up.

12. Glossary

• AI editing agent — The system actor that performs video editing on the creator's uploaded footage in response to the creator's editing direction, and produces the edited result. Not a human participant.
• Content Creator — The sole accepted human persona: the person who produces their own videos and uses hyper-hi to have them edited by the AI agent.
• Editing direction — The creator's stated instruction, submitted alongside a source video, describing how the agent should edit that footage.
• Edited result — The video output produced by the AI editing agent for a project, presented on Results.
• Project — The creator-owned container that binds a source video, an editing direction, and the resulting edited output together, and that persists across sessions.
• Source video — The creator's own uploaded footage, provided to the AI agent as the material to be edited.
• Protected destination — A page reachable only with an established identity: Projects, Upload, and Results.
• Anonymous access surface — A page reachable without an identity: Landing, Sign Up, and Login.