Kontext Media

Capabilities

Kontext Media (/ai-media) makes images, clips, voice, music, presenter videos, reels, promos, motion videos and sites from a film — on your own keys. Views: Studio, Edit, Library, Jobs. Long work runs as server jobs you can leave. Image may use a platform fallback with a banner; keys are billed by the provider, not your AI credits.

• Studio — Create: Image, Video clip, Voiceover, Music · Finished videos: Reel, Promo, Motion video, Avatar explainer, Cinematic UI

• Refine & Finish — Upscale/remove-bg; captions, dub, export presets (each a server job)

• Finished videos — Reel from a recording (with cuts review), Promo, Motion video; the canvas is the style library

• Jobs — Running / Needs review / Failed / Done; cancel, open the result

• Edit — Multitrack timeline; renders with ffmpeg or the HyperFrames render box

• Unfold — a film becomes a page it plays under as you scroll

• Library — Collections, pins and tags saved on your account

• Publish — Download, share link, Network, and social when connected

What is Kontext Media?

Kontext Media (/ai-media) makes images, video clips, voiceovers, music, talking-avatar explainers, reels from a recording, brand promos, motion-graphics videos, and films that become web pages. Views: Studio, Edit, Library, Jobs. Stats: Saved and Keys live.

It runs on your own API keys (Profile → API keys, media). Image can fall back to a platform model with a loud banner. Video, voice, music and avatars need a key. Those keys are billed by the provider — not your AI credit wallet.

Studio rail

Two groups. Create makes one piece from a prompt. Finished videos are made as server jobs; each one fills the canvas and controls with its own screen. Video clips are 4 / 6 / 10 seconds; longer video is a finished video.

Jobs — long work runs on the server

Video clips, captions, dubs, exports, timeline renders, avatar renders, B-roll, Cinematic films, reels, promos, motion videos and Unfold sites run as jobs on Kontext's servers. They keep going when you close the tab, sign out or switch devices.

The Jobs view filters Running, Needs review, Failed, Done, All. Each row says its stage in words (Rendering frame 212/450), the provider, an estimated cost when known, and N of M parts done for multi-part work. Cancel stops a job — and the provider's render where the provider allows it. Open result opens what it made. A failure shows the provider's own error.

Finished jobs are filed into your library the next time Media opens, on any device. Large files go to your cloud storage if you set one, otherwise they stay on Kontext's media engine, which keeps them and lets video seek. A save your account refused is named, never silently dropped.

Reel, Promo and Motion video

Three kinds under Finished videos on the Studio rail. Each fills the canvas with the style library and the controls with its form; the button at the foot of the controls starts a server job that lands in the library as a new item. There is no separate Reels & promos tab.

Reviewing a reel's cuts

With Let me review the cuts on, a reel stops once under Jobs → Needs review. Each candidate is a Repeated word, Retake or False start, with what it removes and keeps. Hear it plays the recording around the cut. Recommended cuts start ticked; Unticked candidates stay in the video. Apply N cuts & continue finishes the reel.

With review off, only the cuts Kontext recommends are applied, without asking.

Edit — the timeline

Lay clips, images and audio on tracks; select a clip to set its transform and keyframes; add text. Render is a server job that returns a new version.

Plain cuts render with ffmpeg. A timeline with text or keyframed motion renders on the motion-graphics render box (HyperFrames). You edit the same way either way. It is a compositing timeline, not a full video editor.

What is Unfold?

Unfold is a Kontext Media way of turning a film into a website: the page plays the film under the scroll wheel, act by act, with your copy on top. Make one from a video in Media (Unfold on the canvas) or from the Website composer (Page · Unfold). A server job builds and deploys it; follow it under Jobs.

Sites from a film — Cinematic UI and Unfold

Cinematic UI films a hero for an ordinary landing page. Film Cinematic UI, then Make this a site opens Workspace with the landing wired to the film.

Unfold makes a page the film plays under as you scroll, your copy on top. On a video, Unfold asks for a Site name, What is the page for? and Deploy when built, then builds and deploys it on the server (usually 10–25 minutes). From a prompt, choose Page · Unfold in the Website composer.

An Unfold page is front end only and its film is hosted on Kontext's media engine. With no server to deploy to, the job ends with the site built and says hosting must be set up once from Deploy.

Avatar explainer (Creator Lab is retired)

Avatar explainer on the rail: write or draft the Script, pick the Avatar provider (HeyGen, D-ID or Synthesia), an Avatar and a Voice, and render — a job; you do not need to wait. Then Handoff (Descript, Google Drive, brief, ZIP, re-import, AI B-roll assist), Polish (trim and reframe) and Distribute.

Creator Content Lab is retired. Its links open the Avatar explainer and its projects are already in your Media library. Send nobody to a Creator Lab desk.

From Business and from Talk

Business → Ways to grow → Make videos opens Studio on Promo with the business named.

Kontext and your staff can make media only when the media module is on their allowlist. kontext.media.produce_video_from starts the same reel, promo or motion-video job and says to follow it under Jobs; kontext.media.styles.list lists the style cards.

Not yet

• No owner screen to edit the brand kit by hand — an assistant sets it.
• Edit is a timeline for compositing — not a full video editor.
• Render times on the live render machine are not measured yet.

Route: /ai-media · OS app ai-media · BYOK generative studio

Kontext Media is Kontext’s owner-facing AI media studio — images, video clips, voiceovers, music, avatar explainers, reels cut from a recording, brand promos, motion-graphics videos, and films that become web pages (Cinematic UI and Unfold) — using your API keys stored on your account (Profile → API keys). Anything long runs as a server job that keeps going when the tab closes and shows in the Jobs view; results land in the library as media projects.

User journeys

KM-01. Open Kontext Media from module switcher

Path: Module switcher → Kontext Media · /ai-media

You do:

1. Sign in with the Kontext chooser (Google, Microsoft, Apple, passkey, Kontext ID, or recovery).
2. Open the global module switcher and choose Kontext Media (or go to /ai-media).
3. Confirm the module header and Studio / Edit / Library tabs load.
4. Check hero stats — Saved count and Keys live (zero until Profile keys are set).

Result: You are in the BYOK media studio with full module chrome.

KM-02. Open from Kontext OS

Path: Kontext OS → Start menu · dock · desktop → Kontext Media (ai-media)

You do:

1. Open /os and launch Kontext Media from the app catalog, desktop icon, or Start menu Tools category.
2. Media opens in an OS window — the module header is hidden and the tabs fill the pane.
3. Or open /ai-media in a browser tab from the module switcher — same studio, standalone chrome.
4. From a piece on the canvas, Open in OS opens its project in the OS when you need files or deploy.

Result: Media is a first-class OS app — not buried under Workspace.

KM-03. Open from Today Media shortcut

Path: My Kontext → Today → Media card · Cockpit dock

You do:

1. Open My Kontext → Today when OS Intelligence is active.
2. Tap the Media shortcut (images, video, voice) on the cockpit dock or three-tabs layout.
3. Confirm /ai-media or the OS ai-media window opens.
4. Return to Today after saving — library count updates on next visit.

Result: Today surfaces Media as a one-tap creative entry without hunting the catalog.

KM-04. Use a deep link

Path: URL: /ai-media?tab=…&kind=…&prompt=…&project=…&flow=…&motif=…

You do:

1. Build or follow a link. tab opens a view: studio, edit, library or jobs.
2. kind picks the rail (image, video, audio, music); kind=avatar opens Avatar explainer; prompt fills the prompt box; project loads that piece on the canvas.
3. flow opens Studio on a finished-video kind — reel, promo or motion — and motif prefills the promo line. Older links that say tab=motion (or flow=styles) still work and open Studio on that kind (Reel when none is named).
4. Open the URL while signed in.

Result: A link lands on the right view, flow and piece — this is how Business, Talk and old Creator Lab links reach Media.

KM-05. Connect media API keys in Profile

Path: Profile → Settings → API keys · /profile/settings?tab=apikeys

You do:

1. Open Profile → Settings → API keys.
2. Add keys for providers you need — e.g. Replicate, Fal, OpenAI, Gemini (image); Runway, Kling, Luma (video); ElevenLabs (voice); Suno (music); HeyGen (avatar path).
3. Save — keys persist on your user canister, not browser storage.
4. Return to /ai-media — Keys live stat and BYOK strip show connected providers.
5. Generate stays disabled until at least one key matches the selected media kind.

Result: BYOK keys unlock Generate for each media type — Kontext never stores keys on platform servers.

KM-06. Choose provider and model

Path: Studio → controls panel → Provider · Model

You do:

1. Pick a media kind on the left rail (Image, Video, Voiceover, Music).
2. If multiple keys exist, open Provider dropdown — only configured providers appear.
3. Select Model when offered (quality vs fast tiers on Replicate/Fal/etc.).
4. For Replicate power users, expand Advanced: custom Replicate model and enter owner/model-slug (KM-12).

Result: Generation routes to your chosen provider and model for that kind.

KM-07. Generate an image

Path: Studio → Image → prompt → Generate Image

You do:

1. Select Image on the create rail.
2. Choose Aspect ratio (1:1, 16:9, 9:16, 4:3).
3. Describe the shot in the prompt box — or tap a quick-start preset (KM-11).
4. Click Generate Image — canvas shows progress, then the image.
5. Project saves automatically; Saved count increments.

Result: A media project with manifest and primary output lands in your library.

KM-08. Generate a video clip

Path: Studio → Video clip → duration → Generate Video

You do:

1. Select Video clip on the create rail.
2. Pick Duration (4s, 6s, 10s) supported by your provider.
3. Enter motion prompt or use a B-roll / trailer preset.
4. Optionally set Cloud storage target before generate (KM-18).
5. Click Generate Video — preview plays in canvas when ready.
6. Use Finish toolbar for captions/dub/export (KM-16).

Result: Short-form video clip is saved with optional remote storage upload.

KM-09. Generate a voiceover

Path: Studio → Voiceover → tone → Generate Audio

You do:

1. Select Voiceover on the create rail.
2. Paste narration script in the prompt area.
3. Pick Voice tone — neutral, warm, or energetic (ElevenLabs mapping).
4. Click Generate Audio — audio player appears on canvas.
5. Download or use in Edit tab timeline (KM-25).

Result: Voiceover file attaches to a media project for podcasts, ads, or explainers.

KM-10. Generate music

Path: Studio → Music → mood → Generate Music

You do:

1. Select Music on the create rail.
2. Describe genre, mood, and use case (demo bed, cinematic, lo-fi, full song).
3. With a Suno key: pick Original, Cover, Mashup, or Instrumental. Cover and mashup need source tracks.
4. Pick Mood chip — appended to the prompt for Suno/Replicate.
5. Generate — preview audio on canvas.

Result: Original song, cover, mashup, or instrumental bed saves to the library for video finishing or download.

KM-11. Use a quick-start preset

Path: Studio create rail → Quick starts chips

You do:

1. Select Image, Video, Voiceover, or Music on the create rail.
2. Scroll Quick starts under the kind cards.
3. Click a chip (e.g. YouTube thumbnail, B-roll, podcast intro, lo-fi focus).
4. Prompt box fills — edit before generating if needed.
5. Click Generate.

Result: Presets jump-start common formats without writing prompts from scratch.

KM-12. Advanced custom Replicate model

Path: Studio → Replicate provider → Advanced: custom Replicate model

You do:

1. Connect a Replicate API key (KM-05).
2. Select Replicate as provider for image, video, or music.
3. Click Advanced: custom Replicate model.
4. Enter full slug owner/model-name.
5. Generate — job uses your custom model instead of catalog default.

Result: Power users can run any Replicate model the platform catalog does not list.

KM-13. Switch output versions

Path: Studio canvas → Versions strip

You do:

1. Open a project with multiple outputs (after refine or regenerate).
2. Locate the version chips below the canvas.
3. Click Original or v2, v3, etc.
4. Preview updates — publish and download use the selected version.

Result: You pick which rendition to ship without losing earlier versions.

KM-14. Regenerate a variation

Path: Canvas toolbar → Variation

You do:

1. With an active project on canvas, click Variation (or Try again after failure).
2. Prompt restores from last manifest when available.
3. Confirm generate — new output appends to version strip.

Result: Another take on the same prompt without leaving the project.

KM-15. Refine an image

Path: Canvas toolbar → Refine

You do:

1. Generate or open an image project on Studio canvas.
2. Click Refine — panel opens with operations.
3. Choose Upscale, Remove background, Variation, Restyle, or Edit region.
4. For Restyle/Edit region, enter a short change prompt.
5. Apply — new version appears on the version strip (KM-13).

Result: Image post-processing runs through your connected provider without re-prompting from scratch.

KM-16. Finish video — captions, dub, export

Path: Canvas toolbar → Finish

You do:

1. Open a video on the Studio canvas and pick the version to finish on the strip.
2. Click Finish — tabs: Captions, Dub, Export.
3. Captions (OpenAI Whisper, needs an OpenAI key): optional language code → Generate captions.
4. Dub (ElevenLabs): pick Spanish, French, German, Portuguese, Hindi, Japanese, Italian or Chinese → Dub video.
5. Export: YouTube 16:9, YT Shorts 9:16, TikTok 9:16, Reels 9:16, LinkedIn 16:9 or Square 1:1 → Export.
6. The panel says Running on the server — you can close this. Each result arrives as a new version; follow it under Jobs.

Result: Captions, dubs and platform cuts run as server jobs — a 10-minute dub survives closing the tab.

KM-17. Download outputs

Path: Publish panel or canvas → Download

You do:

1. Open active project on Studio canvas.
2. Click Download on canvas toolbar — saves selected version.
3. Or open Publish panel → Download / Download all for every output file.
4. File lands in your device downloads folder.

Result: You own the bytes locally regardless of publish path.

KM-18. Configure video cloud storage

Path: Studio → Video controls → Media storage panel

You do:

1. Select Video clip kind — storage panel appears for signed-in users.
2. Pick provider: Google Drive, Box, Dropbox, or OneDrive.
3. Click Connect if Composio OAuth is missing — complete popup auth.
4. Save preference — large video outputs can upload to your cloud folder on generate.

Result: Video assets can mirror to your connected drive instead of only canister storage.

KM-19. Open a piece in the OS

Path: Canvas toolbar → Open in OS

You do:

1. Generate or open a media piece on the Studio canvas.
2. Click Open in OS.
3. Kontext opens the piece's project in an OS window.
4. Use this when you need project files, code or deploy tied to the same project.

Result: Media pieces are real Kontext projects — the studio is their home, the OS is where they meet the rest of your work.

KM-20. Browse the media library

Path: Kontext Media → Library tab

You do:

1. Open /ai-media and click Library.
2. Scan grid of saved projects — thumbnail, kind badge, provider hint.
3. Use collections sidebar: All, Pinned, or custom collections (KM-22).
4. Click a card Open — returns to Studio with that project active.

Result: Every generation appears in Library automatically — no manual save.

KM-21. Search and filter by kind

Path: Library → search · filter chips

You do:

1. Open Library tab.
2. Type in Search your media… — matches title, kind, provider, tags.
3. Tap filter chips: All, Image, Video, Voiceover, Music.
4. Combine with collection sidebar to narrow further.

Result: Find the right asset quickly in large libraries.

KM-22. Create and manage collections

Path: Library sidebar → collections

You do:

1. Open Library → collections sidebar.
2. Click New collection, name it (Campaign Q3, Client A, …).
3. Assign projects from card menu or drag actions when offered.
4. Rename or delete collections from sidebar menu — projects remain; only grouping changes.

Result: Collections organize shoots and campaigns on your canister (media.collections.*).

KM-23. Pin a media project

Path: Library card → pin control

You do:

1. Open Library and locate a project card.
2. Toggle Pin — project appears under Pinned collection filter.
3. Pins sync to user canister (kontext.media.pins.set).
4. Unpin when the hero asset is no longer priority.

Result: Pinned items stay at the top of your organized library view.

KM-24. Tag a media project

Path: Library card → tags editor

You do:

1. Open Library → select a project card tag field.
2. Add comma-separated tags (launch, thumb, podcast-ep-12).
3. Save — tags persist on canister and appear in search (KM-21).

Result: Tags complement collections for flexible filtering.

KM-25. Edit on the timeline

Path: Kontext Media → Edit tab

You do:

1. Generate or open a piece in Studio or Library, then switch to Edit. With nothing selected, Edit asks you to pick from Library.
2. Use Add to timeline to lay clips on tracks; select a clip to edit its transform and keyframes in the inspector.
3. Save — Timeline saved stores the document on the project.
4. Press Render. The panel says Rendering on the server — you can close this.
5. The render lands in your library as a new version; follow it under Jobs.

Result: A non-linear edit renders on the server. Plain cuts go through ffmpeg; a timeline with text or keyframed motion goes through the HyperFrames render box (KM-49).

KM-26. Publish to Kontext Network

Path: Studio → Publish → Kontext Network

You do:

1. Open finished media on Studio canvas.
2. Click Publish → Ship it panel.
3. Under social section, click Post to Kontext Network.
4. Confirm upload — public asset URL is created if needed.
5. Network post opens in new tab; status shows link when live.

Result: Your media becomes a Network post under your handle — ecosystem loop for reach.

KM-27. Share on X

Path: Publish panel → X · Profile X BYOK + Composio

You do:

1. Connect X: save OAuth Client ID/Secret in Profile and connect via Composio (see platform X BYOK guide).
2. Open Publish on a ready image or video.
3. Click X — auto-posts when X is connected.
4. If not connected, browser compose opens with prefilled text and public link.

Result: X share uses platform BYOK path when wired; otherwise safe browser intent fallback.

KM-28. Publish to YouTube or social platforms

Path: Publish panel → YouTube · LinkedIn · TikTok · Instagram

You do:

1. Connect platforms in Integrations / Composio (Business connector status gates buttons).
2. Open Publish on video (or finished export).
3. Click connected platform button — disabled platforms show Connect hint.
4. Wait for upload job — success opens platform URL when returned.

Result: Video ships to connected social accounts from one panel — no re-upload manually.

KM-29. Copy a public share link

Path: Publish panel → Copy link

You do:

1. Open Publish on an active project.
2. Click Copy link — public share URL uploads asset if not already public.
3. Paste link in email, DM, or landing page.
4. Link format includes /ai-media?project= deep link where applicable.

Result: Anyone with the link can view the shared asset per public storage rules.

KM-30. Start an avatar explainer video

Path: Studio → Avatar explainer rail card · /ai-media?kind=avatar

You do:

1. On the Studio rail, click Avatar explainer (Scripted talking video → publish).
2. Pick a quick start (Product intro, Feature explainer, Changelog update) or write the Script — the AI draft button can write it for you.
3. Choose Avatar provider (HeyGen, D-ID or Synthesia), an Avatar, and a Voice (ElevenLabs, HeyGen voice or OpenAI TTS).
4. If a key is missing the panel names it (Add HeyGen in Profile → API keys); Manage keys opens Profile.

Result: The presenter video starts inside Media. There is no separate Creator Lab desk any more.

KM-31. Render an avatar video and distribute it

Path: Avatar explainer → Render → Distribute

You do:

1. With a script, avatar and voice set, start the render.
2. The canvas says Rendering in the background — You don't need to wait here. The render is a server job; it shows under Jobs.
3. When it finishes: Your avatar video is ready to watch.
4. Open Distribute — YouTube, LinkedIn, TikTok, Instagram, a quick X announce, or a Kontext Network post.
5. The video is also a Media item in Library.

Result: A talking-head video from script to post, with the long render off your browser.

KM-32. Generate media via KESA or Kontext

Path: Kontext / OS employee with media module scope · KESA kontext.media.generate

You do:

1. Enable Kontext Ecosystem with media module on an OS employee or agent profile.
2. Ask in Kontext or agent chat: Generate a 16:9 hero image for my launch (or list library / refine).
3. Confirm KESA executes kontext.media.generate or opens /ai-media with open-link result.
4. Complete BYOK and provider steps in Media if the agent reports missing keys.
5. Review output in Library — agent actions respect same canister projects as the studio.

Result: Media is operable through KESA like other platform modules — studio remains source of truth.

KM-33. Follow a job in the Jobs tab

Path: Kontext Media → Jobs · /ai-media?tab=jobs

You do:

1. Start anything long — a video clip, captions, a dub, an export, a render, a reel, a promo, an avatar video, an Unfold site.
2. Open Jobs. Filters: Running, Needs review, Failed, Done, All.
3. Each row shows the stage in words (Rendering frame 212/450, Fetching assets 3/5), the provider, an estimated cost when known, and N of M parts done for multi-part jobs — click it to see every part.
4. A failed job shows the provider's own error text.

Result: One place shows everything Media is doing for you, whether or not the tab that started it is still open.

KM-34. Cancel a job

Path: Jobs → a running row → Cancel

You do:

1. Open Jobs and find the row under Running.
2. Click Cancel.
3. The row moves to Failed as Canceled.
4. Where the vendor supports a cancel, Kontext asks it to stop. Where it does not, the vendor may finish the render it already accepted.

Result: Work you no longer want stops being watched and billed where the vendor allows.

KM-35. Open a finished job's result

Path: Jobs → Done → Open result · Open file

You do:

1. Open Jobs → Done.
2. Open result opens the Media piece the job filed (reels, promos, motion videos and avatar videos arrive as new library items; captions, dubs, exports and renders as new versions of the piece you started from).
3. A job with no project of its own shows Open file instead, which opens the rendered file in a new tab.

Result: From "it finished" to the thing it made in one click.

KM-36. Close the tab and come back later

Path: Any long job → close the browser → return to /ai-media

You do:

1. Start a long job and close the tab, or sign out.
2. Come back later on any device signed in to the same account.
3. Open Jobs — the job is still running, or it is done.
4. Open Library — the next time Media opens on any device, every job that finished while you were away is filed into its project and your media index.

Result: Media work runs on the server, not in a browser tab; the library catches up the moment you come back.

KM-37. When the library cannot be read

Path: Library → Your library could not be loaded → Retry

You do:

1. Open Library. If your workspace could not be read, the tab says Your library could not be loaded instead of showing an empty shelf.
2. Click Retry.
3. If a save of a large file was refused, the piece is marked as not stored and you are told which file — nothing pretends it saved.

Result: "Could not read" never looks like "you have nothing".

KM-38. Film a Cinematic UI landing

Path: Studio → Cinematic UI rail card

You do:

1. On the Studio rail, click Cinematic UI (Film a landing, then deploy the page).
2. Describe what happens as a visitor scrolls, or pick a quick start (Smoothie drop, Watch on the wrist, Pizza in the box, Rain jacket).
3. Pick an image provider and a Film provider (your keys).
4. Click Film Cinematic UI. The hero still and film are made as a recipe on the server.
5. The canvas previews the landing around the film — opening, move in, payoff.

Result: A short hero film and its still, made for a landing page.

KM-39. Make a Cinematic film into a site

Path: Cinematic UI → Make this a site

You do:

1. Film the Cinematic UI first (KM-38). Until then the button says Film it first.
2. Click Make this a site.
3. Pick or confirm a server pair when asked.
4. Workspace opens with the landing already wired to the film; build and deploy it from there.

Result: An ordinary landing page with your film as its hero, built in Workspace.

KM-40. Unfold a film into a site

Path: Studio canvas (a video) → Unfold → Unfold into a site

You do:

1. Open a video on the Studio canvas and pick the version on the strip.
2. Click Unfold. The panel says: Turns this film into a page that unfolds as you scroll: the film plays under the wheel, with your copy on top.
3. Give it a Site name and answer What is the page for? — who it is for, what it offers, the one thing a visitor should do.
4. Leave Deploy when built on to publish it, or off to build only.
5. Click Unfold. Running on the server … Usually 10–25 minutes. Follow it under Jobs.
6. The site appears as a new project in your workspace, with its link when deployed.

Result: A scroll-driven page built around your film — the film scrubs as the visitor scrolls.

KM-41. Start an Unfold page from the Website composer

Path: Landing composer → Website → Page · Unfold

You do:

1. In the prompt composer, choose the Website tab.
2. Under Kind of website, pick Page · Unfold (A page that unfolds as you scroll, built around a film) instead of Page.
3. Describe the page and create it. The choice is remembered on this device.
4. The project is built on the Unfold layout and tagged so later updates stay Unfold.

Result: An Unfold page from a prompt. Add or replace its film from Media with Unfold (KM-40).

KM-42. Follow an Unfold build's stages

Path: Jobs → the Unfold row → N of M parts done

You do:

1. Open Jobs and expand the Unfold job's parts.
2. From a prompt with no film, the stages are: stills for each act → one clip per act → stitch → encode → project → build → deploy → verify.
3. From an existing film (KM-40), the first three are skipped.
4. If your account has no server pair it can claim, the job finishes with the site built and says hosting needs setting up once from the project's Deploy tab.

Result: You can see which stage an Unfold site is in and what, if anything, needs you.

KM-43. Make a reel from a recording

Path: Studio → Finished videos → Reel · /ai-media?flow=reel

You do:

1. In Studio, press Reel on the rail, under Finished videos. The canvas shows the style library; the controls on the right hold the form.
2. Pick a Recording from your library, or …or upload one (up to 500 MB).
3. Name it, and optionally say What should the reel land on?
4. Choose: Let me review the cuts (on by default), Also a 16:9 cut, Use my brand kit, and Transcribe with ElevenLabs Scribe or OpenAI Whisper.
5. Optionally pick a style on the canvas (KM-47). Click Make the reel at the foot of the controls.
6. It runs on the server. With review on it stops once for you (KM-44); otherwise usually 10–20 minutes.

Result: A 9:16 reel — silences and stumbles cut, a hook in the first seconds, word-synced captions — filed in your library as a new item.

KM-44. Review the cuts

Path: Jobs → Needs review → the reel's cut list

You do:

1. Open Jobs → Needs review. The reel lists every cut it found: Repeated word, Retake or False start, with its confidence, what it removes and what it keeps.
2. Press Hear it to play the recording around a cut.
3. Tick the cuts you want. The ones Kontext recommends start ticked; Unticked candidates stay in the video.
4. Click Apply N cuts (X s) & continue.

Result: You decide what leaves your recording; the reel then renders with exactly those cuts.

KM-45. Make a promo

Path: Studio → Finished videos → Promo · /ai-media?flow=promo

You do:

1. In Studio, press Promo on the rail, under Finished videos.
2. Write The brand or the motif, in one line (A paper plane that becomes our logo), and anything else it must say.
3. Pick 16:9, 9:16 or 1:1, and a length from 10 to 30 seconds.
4. Keep Use my brand kit on to use your colours and fonts; optionally pick a style.
5. Click Make the promo — usually 5–15 minutes on the server.

Result: A brand showreel: one motif through six or seven labelled chapters, cut to a beat grid, ending on a held logo lockup.

KM-46. Make a motion video from a brief

Path: Studio → Finished videos → Motion video · /ai-media?flow=motion

You do:

1. In Studio, press Motion video on the rail, under Finished videos.
2. Write the Brief — who it is for, the three points it makes, how it should feel.
3. Pick 16:9, 9:16 or 1:1 and a length from 5 to 120 seconds.
4. Keep or clear Use my brand kit; optionally pick a style.
5. Click Make the video — usually 5–15 minutes.

Result: A motion-graphics video — type, shapes and transitions, no footage needed.

KM-47. Browse the style library and pick a style

Path: Studio → Reel, Promo or Motion video → the canvas

You do:

1. Press Reel, Promo or Motion video on the rail. The canvas is the style library. Filter by style (All styles, then each style with its card count) and by Every purpose / a purpose.
2. Hover a card to play its preview; each card lists its slots and its length range.
3. Click Use this style. The bar above the cards names the pick.
4. The pick carries across Reel, Promo and Motion video; Clear removes it and the composition follows your brand kit.

Result: Reels, promos and motion videos follow a designed style card instead of a blank page.

KM-48. Render a timeline with text or motion

Path: Edit → text track or keyframes → Render

You do:

1. In Edit, add a text clip, or keyframe a clip's position, scale, opacity or rotation.
2. Press Render.
3. Kontext sees the timeline needs more than cuts and renders it through the HyperFrames render box; a timeline of plain cuts renders through ffmpeg.
4. Follow the render under Jobs — the line reads Rendering frame N/M.

Result: Titles and animated moves render the way you placed them, with no change to how you edit.

KM-49. Hand off, polish, or add B-roll to an avatar video

Path: Avatar explainer canvas → Handoff · Polish

You do:

1. Select a finished avatar video on the canvas.
2. Handoff: Send to Descript, make a Google Drive folder, Copy brief, or Download ZIP pack — then Re-import final edit (or Poll Descript) to bring the finished cut back.
3. AI B-roll assist (in Handoff): Suggest beats from script, then Generate a clip per beat. Clips run as jobs.
4. Polish: trim start and end, pick 16:9 YouTube or 9:16 vertical, then Apply polish.

Result: The presenter video can go to an outside editor and come back, or be finished here.

KM-50. Make videos from Business

Path: Kontext Business → a solution → Ways to grow → Make videos

You do:

1. Open a business solution and find Ways to grow.
2. Click Make videos (A promo for your business, a reel from a recording, or a presenter video — made in Kontext Media).
3. Media opens in Studio on Promo with your business named as the motif.
4. Make the promo (KM-45), or switch to Reel or to Avatar explainer for a presenter.

Result: Business videos are made in Media, and land in the same library and Jobs list.

KM-51. Ask Kontext for a reel, promo or motion video

Path: Talk / an agent with the media module → kontext.media.produce_video_from

You do:

1. Make sure the assistant or agent has the media module on its allowlist.
2. Ask, e.g. Make a 20 second promo about our spring launch, Cut a reel from my latest recording, or Make a motion video explaining our three plans.
3. The assistant starts the job and says where to follow it (/ai-media?tab=jobs).
4. It can list styles first (kontext.media.styles.list) and pass a style card.

Result: The same server jobs as Studio's Reel, Promo and Motion video, started by asking.

KM-52. Old Creator Lab links

Path: /creator, /creator/studio and other Creator Lab links

You do:

1. Open an old Creator Lab link or bookmark.
2. You land in Kontext Media → Avatar explainer (/ai-media?kind=avatar).
3. Creator Lab projects you already had appear in the Media Library — they are moved into your media index once, automatically.
4. Public handoff links you shared (/creator/handoff/…) still open.

Result: Creator Content Lab is retired; nothing you made there is lost.

How it works

Route /ai-media · OS app ai-media. Four views — Studio (rail: Create · Finished videos), Edit, Library, Jobs.

• Your keys. Provider keys live on your account (Profile → API keys). A job carries the key it needs sealed and never writes it into the job record. Providers bill you; the no-key image fallback is the one platform-paid exception and says so.
• Server jobs. Video clips, captions, dubs, exports, timeline renders, avatar renders, B-roll, Cinematic films, reels, promos, motion videos and Unfold sites run on Kontext's media engine: queued → running → needs review → done / failed / canceled, with the stage in words and per-part progress. Every stage records what it made, so a server restart resumes rather than re-billing a provider. Closing the tab changes nothing.
• Filing. The next time Media opens (any device), finished jobs are filed into their project: small files onto your account, large ones to connected cloud storage or left on the media engine (kept, seekable). Filing is repeat-safe; a refused save is named.
• Library. Media projects with a manifest of versions; newer workspaces keep a typed media index the library reads first. Unreadable is reported, never shown as empty.
• Render box. Reels, promos, motion videos and timelines with text or keyframes render from HTML compositions on a separate HyperFrames machine (frame-by-frame progress, local fonts and animation library, assets only from Kontext's file host). Compositions are written by Kontext's model under the style library's rules, linted, repaired once, rendered.
• Reels. Word-level transcript (ElevenLabs Scribe / OpenAI Whisper) → silences → mistake candidates (repeated word, retake, false start) → optional owner review → approved cuts applied to the source in one encode → hook + word-synced captions → 1080×1920 (+1920×1080).
• Style library. 406 cards in Paper Collage and Creator Punch, by purpose (label, lower-third, overview, section, stat).
• Unfold. A fixed scroll engine (open-source scroll-craft) plus desktop and phone scrub encodes of the film (dense keyframes for instant seeking); film hosted on the media engine; front-end-only site. From a prompt the job also makes per-act stills and clips with pinned start and end frames (Runway Veo 3.1 or Kling), then stitch → encode → project → build → deploy → live check.
• Avatar. HeyGen / D-ID / Synthesia renders and B-roll are jobs; Creator Content Lab is retired into Media and its links redirect here.
• KESA. kontext.media.* — generate, refine, library, collections, tags, pins, brand kit, styles.list, produce_video_from (reel / promo / motion).

Benefits

• One studio for stills, clips, voice, music, presenters, reels, promos, motion graphics and film-driven pages.
• Long work runs on the server — leave, come back on any device, find it done.
• You approve what gets cut from your own recording.
• Your keys, your library, publish to the Network and connected channels.

Who it's for

Primary: owners and teams making marketing, product and social video. Also: Business owners (Ways to grow → Make videos) and assistants with the media module. Not: a zero-key path for video, voice, music or avatars; not a full video editor.