Wevi

View saved version →

---
name: capture-product-ui
description: Capture a live product page or app screen into a Wevi scene. Open the page in Wevi's browser, move to the right view one step at a time, let the user sign in themselves, pick the components the template's frames are made for, and re-pick when the choice looks wrong. Use when a scene's template needs real UI (requiresUiCapture or scenesNeedingCapture), when the user wants their dashboard, table, chart, form or editor in the video, or when a capture came out wrong.
---

# Capture product UI into a scene

How it works: the page is opened in a real browser, every visible element is measured and listed as a candidate; the chosen elements are cut onto their own layer and the background is recaptured with them hidden, so the animation moves a clean component over an intact page. Framing is automatic: the view zooms and scrolls so the components fill about 60% of the frame, sticky bars and cookie banners are hidden. Use `frame: "current"` to keep the view exactly as browsed, `"full"` for the top of the page.

## Procedure per capture scene

1. Know what the scene must show: its narration plus the template's `slots.guide` and `slots.prefer` (the template's own rubric, with the author's notes in `slots.smartGuide`). When present, `slots.frameShape` is the width over height its frames show whole and `slots.frameSize` how large a frame shows a component, in the page's own pixels: pick pieces near that shape and no smaller than about half that size, or they are cut or blown up.
2. Public page and the shot is near the top: call `wevi_capture_scene` with `url` and `focus` keywords.
3. Otherwise `wevi_open_page`, read the outline, reach the right view, then `wevi_capture_scene` with the `sessionId`.
4. If the pick looks wrong in the numbered screenshot, choose by number or id with `wevi_select_capture`. Re-picking is free. Unsure up front? Call `wevi_capture_scene` with `autoSelect: false` to get the candidates and the numbered screenshot first.
5. `wevi_close_page` when the site is done.

## Moving around the page

One step at a time, reading the outline after each: `wevi_go_to` (one of the app's own routes from the outline: dashboard, reports, settings), `wevi_click` (pass `opensMenu: true` when the click opens a dropdown or menu that must stay open for the capture), `wevi_type`, `wevi_scroll`, `wevi_hover`, `wevi_wait`, `wevi_press_key` (Escape dismisses a popover), `wevi_go_back`. Template notes written for the editor translate directly: "navigate to a page that shows X" means a better url or `wevi_go_to`; "click a component" means choose it by focus keyword, selector or number; "keep popovers off the text" means dismiss it first.

## Picking well

- Choose cards, buttons, inputs, charts or media that carry the scene's message. Avoid full-width sections and paragraphs unless the template is a text template (then pick a headline first).
- Thin page? Stay on the product's own domain and look further: first `wevi_scroll` the same page (the best cards, tables and widgets are often below the fold; the outline's sections say where), then try another page of the same site (pricing, features, docs, a dashboard view) with `wevi_go_to`. Never switch to another domain for the product's UI.
- Capturing fewer than the template's maximum slots is fine; fewer than its minimum is not, and a weak pick is worse than scrolling or a second page.
- `fullPageBackground` templates take the whole page and need no picks. A screenshot the user shares at a public URL can fill them through `wevi_capture_from_image`; element templates still need the live page.

## Private apps

Every outline carries `authWall` and `loginUrl`. When `authWall.detected` is true, or `wevi_capture_scene` returns `loginRequired`, send the user `loginUrl` so they sign in themselves in that same browser (SSO and 2FA work), then call `wevi_wait_for_login` with the `sessionId`: it returns when they confirm or the sign-in screen goes away, with the signed-in page's outline and routes. Then `wevi_go_to` the route the scene needs and capture there. Never ask for a password and never type one. A sign-in screen never costs a credit.

## Guardrails

Steps that look like payments, purchases, deletions or account changes are refused; 150 steps per session; only http(s) pages. Browser time is free for the first 5 minutes of a session, then costs 1 credit per started 5 minutes, so close sessions when done.

View saved version →

---
name: product-video
description: Make a short marketing video for a software product with Wevi. Plan a storyboard from the product's own pages, build one project from published motion templates, capture the product's real interface into the scenes that need it, render with music and an AI voiceover, review the result and share the MP4. Use when the user asks for a product video, launch video, feature explainer, product demo clip, social promo or call-to-action outro for their product or website.
---

# Make a product video with Wevi

Wevi builds short marketing videos (3 to 6 scenes, 15 to 45 seconds) from published motion templates. Each scene is one template whose text, colors, logo and image slots are filled for the product; scenes that show real product UI are captured from a live page. Wevi renders every scene in the cloud and joins them into one MP4 with background music and an AI voiceover.

## Before building, ask what is missing

- The product's website or domain: the logo, colors and fonts are read from it. With no website, ask for a public logo image URL.
- The goal of the video (launch, feature explainer, social promo, call to action), the audience and the tone.
- Length or number of scenes, the key messages, whether there should be a voiceover, and brand colors if the user has them.
- Export quality, asked once before the first render and only when the account has paid access (`paidAccess` in `wevi_get_credits`): 1080p is the default; offer 4K only when the user wants it or the video is for a large screen. Free accounts export at 720p with a watermark, so do not ask there.

## Steps

1. Open the product page with `wevi_open_page`. The outline lists `productUi` (the interface on screen), `appEntry` and `authWall`. A product video should show the product's real interface, which is usually behind a login: when `appEntry.looksLikeMarketing` is true and candidates exist, offer to capture inside the app, send the user `loginUrl`, and call `wevi_wait_for_login` with the `sessionId`. Use public marketing pages only when there is no app to reach, the user declines, or the scene is about the website itself.
2. Plan with `wevi_generate_storyboard`, passing the `sessionId`, so the plan uses templates built for the interface actually on that page. Show the plan, adjust it with the user, then pass `createPayload` to `wevi_create_project`: all scenes in one project, with `brand: { websiteUrl }` (or `logoUrl`). Never one project per scene.
3. For each scene in `scenesNeedingCapture`, reach the view that shows that scene's message (`wevi_go_to`, `wevi_click`, `wevi_scroll`), then call `wevi_capture_scene` with the `sessionId`. Check the numbered screenshot it returns; if the pick looks wrong, re-pick with `wevi_select_capture` (free). The `capture-product-ui` skill has the full procedure.
4. Fix every issue listed in `planReview` with `wevi_update_scene`, then `wevi_trigger_render` with the chosen quality and poll `wevi_get_render_status` until `phase` is `completed`.
5. Read `review` on the render status: it lists what a look at the rendered scenes found (a tiny or blurry logo, cut-off or overlapping text, placeholder copy, a broken layout) with a suggested fix per scene. Repair with `wevi_update_scene` and render again; only changed scenes re-render. Share `videoUrl` when `review.status` is `done` and it lists no issues; if an issue cannot be fixed, tell the user what it is.
6. Call `wevi_close_page` when the site is done.

## Revising

`wevi_update_scene` changes one scene's copy, parameters or narration. `wevi_add_scene`, `wevi_remove_scene`, `wevi_reorder_scenes` and `wevi_swap_template` revise the project without rebuilding it. Each scene's `templateCopy` lists text layers still showing the template's own words; set those when the copy matters. `wevi_list_music` and `wevi_list_voices` give the choices for `musicTrackId` and `voiceId` on `wevi_trigger_render` (paid access only); otherwise Wevi picks a track for the brand's tone and the default narrator. `wevi_get_scene_previews` shows single rendered scenes.

## Rules

- One planned story, never a list of features. Open with what the product is and who it is for, or with the viewer's problem and the product by scene 2. Close with a call to action that names the product. Never open cold on a feature claim.
- A different template for every scene; only plain text title templates may return. All scenes in a video share one aspect ratio.
- Narration per scene fits `voiceoverBudget` (about 2.4 words per second, complete sentences). On-screen titles are 6 words or fewer; one idea per scene; narration says what the visual shows.
- Leave the animation IN and OUT parameters unset unless the user asks for a specific move: cuts between scenes are matched automatically.
- Never ask the user for a password and never type one. Sign-in happens in the user's own browser session through `loginUrl` and `wevi_wait_for_login`.
- Check `wevi_get_credits` before large batches. When an action needs more credits than the account has, say how many it needs and how many are available; credits are managed in the user's Wevi account.
- Outside Wevi's reach: free-form or generative footage (people, characters, animals, scenery), editing the user's own video files, changing a template's animation or layout, mixing aspect ratios, vertical video, cloning a voice. Say so plainly and offer what Wevi can do instead.