UGC Ads

1. UGC Ads

A UGC ad is a short, creator-style ad for something you sell: a product, a course, an event. The planner writes it as one to ten short scenes, each with a beat (Hook, Problem, Reveal, Demo, Benefit, Proof, Twist, Call to action) and each showing product b-roll, a creator talking to camera, or the creator with the product. You refine the plan in chat, then render it on Aificient Cloud.

Where: the UGC mode on the creation home screen.

Each scene is rendered by a reference-to-video model (Seedance) from your photos, the creator's portrait, and the voice's own reading of the scene's line, so the creator lip-syncs to exactly those words. The clips are then captioned (if you want), given any on-screen texts you add, and stitched into the final ad.

Crazy hook is a switch inside UGC mode, not a separate mode: the ad opens on something impossible, an absurd event filmed like real footage, that reveals what you sell. A crazy hook is still a UGC ad, with the same project list (UGC ads), the same board, and the same sidebar. Only its name (Crazy Hook), its planning controls, and its lengths change.

UGC mode is available on the desktop, on the web, and on a phone, where the composer's knobs move into the Options sheet. If the server does not offer it yet, the chat says UGC ads aren't available on this server yet. If rendering is not available, it says UGC rendering isn't available right now. Try again later.

Creating a UGC Ad

1. Pick UGC in the mode picker on the creation home screen.
2. Describe what you sell, who it is for, and the offer. Or paste the product's page in the Assets panel and let the app read it.
3. Attach up to 6 photos with the Assets pill, and mark the real places to film in as locations.
4. Pick an Angle, or switch on Crazy hook and pick a Hook idea.
5. Set the format, length, language, and camera feel in the settings pill.
6. Pin up to 2 Characters and pick a Voice.
7. Send the brief. The plan card appears in the chat (see "The Ad Plan").

The Home Screen

The hero reads Turn what you sell into a creator-style ad., with the subtitle Describe your product or service and the offer, attach a few photos, and pick a creator and a voice. I'll write a short scene-by-scene ad you can refine in chat before rendering — or switch on Crazy hook to open it on something impossible. The placeholder is Describe what you sell — a product, a course, an event — who it's for, and the offer….

With Crazy hook on, the eyebrow reads UGC · Crazy hook, the title Open your ad on something impossible., and the placeholder What you sell and who it's for — we'll invent the chaos….

The pill row, left to right: Assets, Angle (or Hook idea with Crazy hook on), the settings pill, Characters, Voice, Skills, and the Crazy hook switch. There is no Find inspiration showcase and no credit chip in this mode.

Photos and Locations

The Assets pill (n/6 badge) holds the photos the ad is built from. Its tooltip spells out the rule: Up to 6 photos, no people: clear shots of what you sell (a product keeps its label, colours and packaging) and of the real places to film in — mark those as locations. Creators come from your characters.

• Up to 6 photos: of what you sell and of the real places to film in.
• No people. Everyone on camera comes from your characters.
• PNG or JPG only, up to 20 MB each.
• Product photos and locations share the 6 slots; marking one as a location costs nothing extra.
• When the plan turns out to sell a service, the hint changes to Optional — up to 6 photos, no people: the venue or the classroom (mark those as locations) and the materials, like a workbook or the course on a laptop. The teacher or host comes from your characters.

The panel (Assets (2/6 · 1 location)) has an Add button, a drop target (Drop PNG / JPG anywhere, or click to browse, N more · up to 20MB each), and the staged photos. Each photo shows a thumbnail, its name, a Location toggle (map-pin icon), and a remove ×.

The Location toggle marks a photo as a place:

• Use as a location: a place the ad is filmed in, not a product. marks it.
• A location: a place the ad is filmed in. Click to use it as a product photo again. undoes it.
• In the tray above the prompt a marked photo carries a Location strip, and the sent message labels it <name> · location.

You can also drop photos anywhere on the home screen (Add photos / Up to 6 photos of the product and its places, no people · N more fit.).

Photo Messages

Importing a Product Page

1. Under the photo list, paste the product's own page into Product URL (for example https://shop.com/product). The hint reads Paste the product's page: its name, details and photos are read for the ad.
2. Click Import. It reads Reading… while it works.
3. The page's name, brand, price, description, key points, offer, and photos are read. The AI picks the photos that clearly show the product, and they fill whatever room is left of the 6.

After an import the product shows twice: as a card in the Assets panel, and as a chip above the prompt with a fanned photo deck. The card shows the main photo, the name, brand · price · site, the description, the key points, the offer, and the other photos, with a × (Remove the product and its photos). Click either to open the product.

• While a product is staged, photos cannot be marked as locations.
• With a product staged and no plan yet, you can send without typing anything; the chat shows An ad for <product name>.

Import Notices

Import Errors

Imports are limited to 20 per hour per account; past that, wait a little and try again.

The Product Modal

Clicking the imported product opens Product, with the site as subtitle.

Each section carries a hint:

• Photos · X of Y — Every photo found on the page. The AI picked the ones marked; click a photo to add it to the ad or take it out. The first one is the main shot.
• Product — What the ad calls it, and what it costs.
• Description — One or two plain sentences: what it is, who it's for, the problem it solves.
• Key points · n/5 — The facts the ad leans on. Short phrases sell best.

A footnote reads Read from <site>. The ad is written from these facts only: no price or offer that isn't here, and no links. The footer has Remove product, Cancel, and Save.

Angle

The Angle pill (ads only) chooses the approach. Once a plan has an angle the pill shows it (… From the plan.), and you can change it with any correction.

Crazy Hook and Hook Idea

The Crazy hook switch (comic-burst glyph) reads Open on something impossible: an absurd event, filmed like real footage, that reveals what you sell. It hides Angle and Camera feel, shows Hook idea, and switches the lengths to 10, 15, 20, or 30 seconds. Flipping the switch on a correction switches the existing plan's format.

Hook idea offers:

• Surprise me — the default; the planner invents the event.
• Delivery chaos, Falls from the sky, Giant product, Force of nature, Animal heist, Explosive reveal, Impossible physics, Trick shot, Chain reaction, Hidden-camera prank, and Time warp — each with a one-line hint.

Format, Length, Language, and Camera Feel

The settings pill (9:16 · 30s · Auto) holds four settings. All four lock once the first plan exists.

• Aspect ratio — 9:16 (default), 1:1, or 16:9.
• Target duration — 15 seconds to 1 minute in 5-second steps (30 by default). A crazy hook offers 10, 15, 20, or 30 (15 by default).
• Narration language — the language the lines are spoken in; Auto reads it from your brief.
• Camera feel (ads only) — see the table.

Creators and Voice

• Characters — Pin up to 2 creators from your library. The first one leads and speaks on camera. The first chip carries a Lead tag. Pinning is optional: with nobody pinned, the planner invents a creator (shown on the plan as New creator). More than two: At most 2 characters can appear in one ad — the first one leads.
• Voice — the voice every line is spoken in, picked from the usual voice picker. By default it is the lead character's voice, then the plan's. Its tooltip says where the current one comes from: Voice: <name>, from <lead name>., Voice: <name>, from the plan., or Voice: none picked — the lead's voice, or your default voice.

Sending the Brief

Send is enabled once you describe what you sell, or a product page is staged. While the planner works, the chat shows Writing the ad (Inventing the hook for a crazy hook), and the send button becomes a stop button.

If something blocks sending, the button's tooltip says what:

The Ad Plan

The reply reads Here's the ad plan. Refine it in chat, or generate the video when it looks right. (Here's the crazy hook. …), followed by the plan card. The card is read-only and shows:

• An Ad or Crazy Hook eyebrow and the title, with chips for the angle, the total length, the format, and the dialogue language (flag and code).
• For a crazy hook, a Hook block: the hook idea, the event, Reveal …, and Twist ….
• The product: up to four photos (a sparkle marks one that will be generated), the name (Untitled product / Untitled service if none), the offer and call to action, and a Details toggle for the description and key points.
• Who and where: the creators (or No one on camera), the voice with a preview button (or No dialogue), and the locations with thumbnails.
• Scenes · N with the total length. Each row shows its number, beat, location, the spoken line in quotes (or what happens), a sound cue on crazy hooks, and its seconds. N images are generated when the project is prepared. notes images that do not exist yet.
• A footer with Video ≈ X cr (Estimated from the scene lengths at the default quality. You choose quality, resolution and captions next.) and the Generate button (clapperboard). Older cards in the chat have Generate disabled.

Keep typing to refine the plan; the placeholder reads Refine the ad: punchier hook, add a demo shot, shorter lines, a different offer… (Refine the hook: a bigger event, a faster reveal, a different twist, a louder reaction…). A correction can bring new photos or locations, a new cast or voice, a new angle or hook idea, the Crazy hook switch, or a newly imported product page, always with some typed text. The reply reads Updated. The ad plan reflects your latest direction.

Warning: If new photos would push past 6, the composer warns first: The ad keeps at most 6 photos — sending these drops one from the plan (location photos first, the oldest first).

Generate checks the plan first and refuses when something is off:

Video Settings for a UGC Ad

Generate slides in Video settings, with the plan beside it on large screens under The plan. The back arrow, Back to chat, returns to the chat.

The button reads Start generation (or Create storyboards with Generate storyboard only on), with a summary such as 9:16 · 30s · Lite · 720p · Captions · ≈ X cr. There is no runtime choice: UGC ads render on Aificient Cloud.

Cost: The estimate covers the video plus one storyboard per scene (or only the storyboards). Generated reference images and the reading of the lines are charged separately.

Starting creates the project (it appears under UGC ads in the sidebar and opens on its board) and starts its first run at once. See "Generating and Resuming a UGC Ad".

The UGC Board

The board runs left to right: the project card, the characters, the Assets card, the scene cards, their storyboards, their clips, the finished cuts (only when captions are on or some scene has texts), the transitions (with two or more scenes), the Stitch node, and the Output. The canvas selects; a modal edits. Click a scene card to open the scene editor, or a character or asset tile to open the asset modal. The Edit with AI bar sits in the bottom-right corner.

Project Card

Project, the ad's title, <N>s · <N> scenes, and under a Style heading the kind and product (UGC ad · <product> or Crazy Hook · <product>).

Character Cards

One per person on camera, the lead first (at most 2). Each shows the portrait (click for full screen), the name with a source chip, the scenes they are in (and N photos beside them when the person has two pictures), and a description. The lead's card also shows the ad's voice: a preview button, the voice's name, and its gender and accent. The voice is changed from the lead's Edit modal (see "Voice and Spoken Lines").

Actions

• Edit — opens the asset modal on that person.
• Add character — a dashed button under the last card (Add a character: from your gallery, from a photo, or invented).

With nobody on camera, a Narrator card takes their place: Anonymous narrator / Nobody is on camera; the lines are spoken off screen., with the voice, an Edit voice button that opens the voice modal, and the same Add character button.

Assets Card

Assets with a count of the images that are not people (click the word to open the sidebar's Assets tab). The tiles run product photos, then other assets, then locations. Empty, the card reads Product photos, assets, places… / Upload a photo or create one with AI.; while the run draws images it glows and reads Generating references · <done> of <total>.

Actions

• Upload button — Upload a product photo, or Upload a photo of the service.
• Atom button — Create an asset with AI.
• Click a tile to open the asset modal (see "Assets, Characters, and Locations").

Scene Cards

The cover is the scene's location photo (or the place in words, No specific place, or The place's photo is not generated yet). On top: the scene number, the beat, and the length. Along the bottom edge: round avatars of the images the scene sends (@Image2 · <name>).

Below the cover:

• The title and the place.
• For a speaking scene: the speech label (On camera, Voice-over, Filmer reacts), the line in quotes, and its voice-line row (see "Voice and Spoken Lines").
• What happens on screen, or Ambient shot.

The footer shows the camera and the sound cue, plus:

A scene that needs review has a red border and a Needs review · <reason> strip (see "Needs Review").

Actions

• Click the card (or press Enter) to edit it (see "Editing a Scene").

Storyboard Nodes

Scene N · storyboard: four quick sketches of the shot (opening frame, two key moments, final frame) so you can check it before paying for its video. Click the sheet to see it full screen.

Actions

• Storyboard / Redraw — shows its price (· 0.6 cr, for example) and draws only that scene: references first if needed, no voice, no video. Hidden while a run is active.

Clip Nodes

Scene N · clip shows the rendered take at its own proportions, with play and seek, full screen (Download, Share, Close), and thumbs up/down once ready.

Under the preview, two things can appear once a take exists:

• A Trimmed chip with the saved trim (−0.4s start · −0.6s end). Its tooltip reads Trimmed in the final video; the clip stays whole. Click to change or remove it., and clicking it reopens the trim modal.
• A grey strip, The read slipped: montaña → montañaña · +la · −siempre, when the delivered clip did not say the line word for word (see "The Read Check").

Actions (all hidden during a run)

• Render / Regenerate — see "Rendering a Scene".
• Scissors — Trim the start or the end for the final video; the clip stays whole (see "Trimming a Clip").
• Bin — Deletes this clip, its finished cut and the final video. Resume Generation renders it again. It asks first.

Trimming a Clip

A trim cuts the start or the end of a scene's clip in the final video only. The clip itself stays whole, nothing renders, and you can undo it at any time.

1. On the clip node, click the scissors (Trim the clip of scene N). The Trim clip modal opens, subtitled Scene N.
2. Drag the handle at either end of the filmstrip (Trim the start, Trim the end); the cut parts dim. The preview loops only the kept part (Play the kept part / Pause, with a 1.2 / 4.2 s counter), and a click on the strip moves the playhead.
3. Read Keeps 4.20 s of 5.00 s and the chips −0.40 s start / −0.60 s end; Reset clears both.
4. Click Save trim (Only the final video is cut again; nothing renders). Once a trim is saved, resetting it and saving reads Remove trim.

The note in the modal sums it up: The clip stays whole — only the final video is cut, and you can undo it any time. Transitions, edge fades, and captions apply to the trimmed span, and a trim survives later edits by the planner or Edit with AI. During a run the button is locked (Wait for the generation to finish); a failed save reads The trim could not be saved.

The Read Check

When a clip lands, the service transcribes it and compares it with the line. If a word differs, the clip node shows a grey strip under the preview: The read slipped: <heard as> → <line> · +<extra word> · −<missing word> (the first three mismatches, then (+N)). Its tooltip reads The clip was transcribed and compared with the line. <all mismatches>. Regenerate to get another read.

It is a warning only: nothing re-renders and nothing is charged. Regenerate the scene for another take, or keep it if the slip does not matter. The transcriber can raise a false alarm, so listen before you spend. Edit with AI sees the slip too (READ SLIPPED in the clip) and can offer to render the scene again. The strip is hidden while a new render runs.

Finished Cuts

Scene N · finished is the clip with its captions and texts burned in.

Transitions, Stitch, and Output

As in a story (see "Scene Transitions"): a join node between each pair of scenes with Change transition, then Stitch (Waiting…, Stitching…, Stitched, Failed) and Output with the final video and Download (full screen Final Video).

Voice and Spoken Lines

Every spoken line is read first by the ad's voice, as its own take. The take is sent with the scene's render, so the clip lip-syncs to it word for word, and the scene lasts as long as its line: the take's length plus a short tail, rounded up to whole seconds, between 4 and 15. A take is dropped when its line, the voice, or the language changes.

Note: A take can be at most about 14.5 seconds (A spoken line runs longer than one clip can carry. Shorten it.).

The Voice Row

The voice row on the lead's card (or on the Narrator card) is read-only: a preview button, the voice's name (or No voice yet), and its gender and accent. An Auto tag marks a voice the app picked for the language, gender, and accent.

Changing the Ad's Voice

1. Click Edit on the lead's card (or Edit voice on the Narrator card). The modal's Voice section reads Every line this character says is read in this voice. (the voice modal: The voice every spoken line is read in).
2. Click Change voice (or Pick a voice; tooltip Changing the voice reads every line again and re-renders every speaking scene) and pick a voice.
3. Read the amber confirmation: <new> replaces <old> for every spoken line., what it removes (every speaking scene's clip and the final video, or No clip uses the voice yet.), and Every spoken line is read again in the new voice; each scene shows its line while it is read.
4. Click Change and read the lines (or Cancel). The lines of every speaking scene that has no clip yet are read again.

The voice modal also has a Spoken lines section with Read / Read again (Read the lines again. Clips already rendered keep the read they were rendered from.), and a Done button. A voice change keeps the speech pace.

Speech Pace

How fast every line is read is set in the sidebar's Settings tab, under Voice: Speech pace with Slow, Med−, Med, Med+ (the default), and Fast (How fast every line is read. Changing it reads them again.). It is not staged with Apply changes: picking a pace opens Change speech pace (for example Read at a slow pace) right away.

• The dialog says what happens: Every spoken line is read again at a slow pace. This deletes the clips of scenes 1, 3 and the final video. … — or … No clip uses the voice yet.
• Confirm with Change and read the lines (Change pace when nothing speaks), or Cancel.
• The control is disabled with The ad has no voice yet or Wait for the generation to finish; a failure reads The pace could not be changed.

The Line Row

The line row on each speaking scene card plays the take (play/pause, seek, time), or reads Not read yet, Reading the line… (the card is outlined in blue), or the error in red. Its ↻ button (Read the line / Read the line again) always reads the line again.

Warning: If the scene already has a clip, reading again warns The clip was rendered from the old read, so it is removed and has to be rendered again.

Messages You May See

Only a missing voice blocks the video; lines that are not read yet are read by the run itself.

Editing a Scene

Clicking a scene card on the board opens the scene editor: Scene N, subtitled with the scene's length, speech, and beat (5 s · On camera · Hook), with Previous scene / Next scene arrows and an N / total counter. (A row in the sidebar's Scenes tab only centres the board on the card.)

Warning: There is no unsaved-changes guard. Closing the editor or stepping to another scene drops the draft; switching tabs keeps it.

The Scene tab

• The title is edited in place at the top (placeholder Scene N); on a crazy hook there is also a Beat.
• What they say — a Who speaks control (On camera, Voice-over, Off camera, or None) and the line box, with the take's player and a hint such as 4.2 s read · needs 5 s, 12 words · needs ~6 s, or Not read yet. A silent scene reads Nobody speaks: the scene plays with its own sound.
• Shows (Creator / Product / Both; Experience for a service) and Camera (see the table), side by side.
• What happens — the action, with a sound strip: Add a sound cue opens a field (A sound, in English — e.g. ice clinking in a glass; Replaces the room tone), and an × removes it (Remove the sound cue).
• The bottom bar — Duration, a −/+ stepper (Between N and 15 seconds; the minimum is what the line needs, at least 4 seconds); Before the line, 0–3 seconds of action before the line starts (Seconds of action before the line starts), for example a sip before the words; and, from scene 2 on, Comes in, a thumbnail of the transition that previews on hover and opens the transition picker.

Note: Before the line is part of the prompt and the duration, so changing it re-renders the clip, and the minimum duration grows by it.

The Location tab

Pick one of the ad's places, No specific place, or New location… (a name and a description; Its photo is generated from this description when you save.). New location… is disabled with No room for its photo when the photo room is full. Below it sits Where in the place, and how (stance first) (for example Sitting on the chair at the table, legs under it; close to the lens).

The References tab

The images the scene sends (n of 5), as chips with their @1, @2 tags; click one to send it or leave it out.

An image you add that the scene's text does not mention yet shows a New in this scene card: Not in the prompt yet — say where it appears, or let the AI decide. Needed before saving. (required for an asset), with Write in (the AI places it) and Write it myself. Until then, saving is blocked with Say how <name> appears in the scene first (References).

The Prompt tab

The prompt the clip renders from, with an Automatic / Edited pill (the tab badge reads custom), Edit / Done, and Rewrite with AI. @Image tags show as picture chips and @Audio1 as the voice; tags for images the scene no longer sends are red (Removed image).

• Rewrite with AI asks What should change? (Every @Image and the voice are kept); the sentences you did not ask to change are kept as they are. On success: Rewritten. Save the scene to render from it.
• An edited prompt says Edited, so changes in the other tabs won't update it., with Undo edits.

The Texts tab

Up to 3 on-screen texts per scene, each up to 120 characters and 3 lines, with a live preview in the ad's format. Empty, the tab reads No text on this scene. Texts are burned into the finished video by the app, so changing them never re-renders the clip.

Emoji are left out (Emoji can't be burned into the video, so they're left out.). The editor also warns about characters the font cannot draw and words wider than the frame.

Footer Buttons

• Render clip / Regenerate clip — with its price; reads Rendering n% or Queued while it renders (see "Rendering a Scene").
• Edit with AI — Describe a change and the AI rewrites the whole scene. It docks a one-line composer (Tell the AI what to change — the whole scene follows; Enter sends, Esc closes) that changes the draft only, keeping line, shot, place, and length in sync, and reports what changed (Changed line, location, length, with Undo).
• Discard — Back to the saved scene; shown only once there are unsaved changes.
• Save scene — saves and says what it did (see the table below).

Save Notices

A notice is sometimes followed by The lines will be read again in the new voice. or Its storyboard no longer matched and was removed — redraw it to check the new shot., and by an action: Render it now or Finish the video.

Warning: Saving a scene that is rendering asks first: Scene N is rendering. If this change affects its clip, that take is discarded when it lands — its credits are still spent. Choose Keep editing or Save anyway.

Assets, Characters, and Locations

Clicking a tile, a character's Edit, or a row of the sidebar's Assets tab opens the asset modal. Its top shows the picture (View full screen), tags, Used in, and the actions; its footer has Delete … and Done.

Edits that do not touch a rendered clip save at once (Saved.). Edits that would remove rendered clips, or waste a take still rendering, ask first in an amber box with Keep it and the confirm button, spelling out the cost, for example:

• This deletes the clips of scenes 1, 3 and the final video. Those clips render again on the next generation.
• The take rendering now for scene 2 is discarded when it lands — its credits are still spent.

A Picture

A person, a product photo, an asset, or a loose place photo. The subtitle says what it is (A character from your library, A person invented for this ad, A photo of the product, An asset the ad shows — not the product, A location photo). The tags give its kind (Person, Product / Service photo, Asset, Location photo) and origin (Your photo, Library, Generated, Edited photo, Not generated).

Actions

• Generate image / Regenerate image — generated images only (Draw this image (one image generation)). Regenerating one that clips use asks Generate a new image?. Your own photos are never regenerated.
• Replace photo — your uploads (Use another photo of yours (PNG or JPG); every scene that sends this one renders again); confirm with Replace the photo?.
• Change person — a library character (Pick another character from your library; every scene with this person renders again). Opens Replace <name> with your gallery.
• Details — the Name; for an asset, What it is — the video reads this, written by AI from the picture and written again whenever the picture changes (a new image or a replaced photo; a name you set is kept); for a generated image, its Prompt. Saving details redraws nothing (The image still shows the previous prompt — regenerate it to match.).
• Scenes — a card per scene: click to send the image with that scene or leave it out (A scene sends at most 5 images; changing what a rendered scene sends renders it again.). Adding an asset to a scene opens that scene's editor so you can say how it appears.
• Delete asset — Delete this asset?. The image leaves the assets and every scene that sent it; those scenes turn red until you check their prompt (see "Needs Review"). A person whose last picture this was leaves the cast.

A Location

A place the ad happens in, tagged Location and No photo, Your photo, Generated photo, or Edited photo. A location needs a description or a photo. A place always keeps one photo; there is no "remove photo".

Actions

• Generate photo / Regenerate photo (or Generate instead over your photo) — draws it from its description.
• Use my photo — uses yours.
• Details — the Name and Description (What every scene here says about the place. Re-describing re-renders their clips.).
• Photos the ad already holds — use a loose place photo for this place (Use for <place>).
• Scenes here — move scenes in or out (Move scene N here?).
• Delete location — Delete this place; its scenes wait for review.

Adding

Add to the assets (Something the ad shows, or a place it happens in) has Assets and Locations tabs; under them, the ways are cards: Generate (Drawn from your words for an asset, Drawn from a description for a place) or Upload (Your photo, as is or edited / Your photo of the place). Assets open on Generate; the way you pick is kept when you switch tabs. An asset is named and described by the AI. An upload can carry an optional Change it with AI (optional) instruction, which produces an Edited photo.

• Buttons: Add & generate, Add & edit, or Add photo.
• While a generated asset is being made, its modal reads Drawing the image — AI names and describes it next., then AI is naming and describing it from the picture…; its Name and description fields wait with Named by AI once the image is ready… / Written by AI once the image is ready….
• Add a character (A person on camera — from your gallery, from a photo, or invented) offers Gallery (One of your characters, the default), Upload (Your photo as their portrait, optionally changed by AI), or Generate (Invented from a description, with Who they are), plus Name (optional) and Gender (Any / Female / Male).
• A generated creator is drawn full-body on a plain white background. A creator made from your own photo keeps that face: if the video provider refuses it as a real person, the service registers the portrait with the provider and submits the scene again by itself; only when that fails do you see a message (see "Render Messages").
• The lead's modal also has the Voice section where the ad's voice is changed (see "Changing the Ad's Voice").
• From the gallery, confirm Add this person? — <name> joins the cast. Add them to a scene from the Scenes tab — no clip changes until then.

Note: Nothing you add joins a scene by itself.

Needs Review

When you delete an asset or a location that scenes use, those scenes are flagged instead of silently changed. A scene is also flagged when its hand-edited prompt still names an image it no longer sends.

• The scene card gets a red border and Needs review · <reason>.
• The sidebar row gets a red number and Needs review.
• The editor shows a red box with the reason and Fix the prompt and save the scene — no video renders until then.

The reasons read like <Name> was deleted from the assets — make sure the prompt no longer counts on it, The location "<place>" was deleted — pick another place and check the prompt, or The prompt still names an image this scene no longer sends — remove it or reword that part.

To clear the flag, do one of these:

• Click Open prompt, fix the prompt, and save the scene. Saving clears the flag.
• Click Looks right (The scene is right as it is — clears the flag, keeps its clip).

Warning: While any scene needs review, no video renders: Scene 3 needs review (an asset it used was deleted).

Rendering a Scene

The clip node's Render (Render this scene on Aificient Cloud) and Regenerate (Render a new take of this scene on Aificient Cloud; the final video is rebuilt when it lands), and the editor's Render clip / Regenerate clip, render that one scene.

1. Click Render or Regenerate.
2. On an ad saved on Lite or Pro, even a first render asks (Render this video? — Choose the quality for this scene.). Pick a quality for just this scene: the ad's own (Like the rest of the ad) or a higher one — Pro (Smoother motion — Steadier, more natural movement than Lite. Just this scene, a bit pricier.) or Max (For a complex scene — More natural movement when there's a lot going on. Just this scene, the priciest.), each with its estimate. With a higher tier the button becomes Render with Pro / Regenerate with Max, and so on. On an ad saved on Max, a first render starts without asking.
3. Confirm. A Pro or Max take renders at the ad's own resolution (720p on a Lite ad) and leaves the ad's saved settings alone.

Warning: Regenerating always asks Regenerate this video? — The current video is replaced by a new render. The old take can't be brought back.

Cost: The dialog shows the cost before you confirm (The new render costs ≈ X credits.).

A render is blocked while something it needs is missing; the button's tooltip says what, ending in No video until then.

When the clip lands, the read check compares it with the line and warns if a word slipped (see "The Read Check"); trimming the start or the end for the final video lives on the clip node too (see "Trimming a Clip").

The UGC Sidebar

The header shows the title, a line such as UGC ad · 5 scenes · 32s · 9:16, and a reload button (Clear app media caches and reload previews). Its tabs are Scenes, Assets, and Settings.

Scenes

The outline of the ad:

• Project (with the product); on a crazy hook, the hook concept with Twist · ….
• Assets (6 images · 2 locations · voice).
• One row per scene: its number, title, status, and place. Click it to centre the board on the scene card; the number badge's border carries the status (green when done, a blue glow while rendering, dashed and pulsing while queued, red when failed).
• Final video — Stitching…, Ready, <done> of <n> clips, or Not stitched.

Assets

Four categories; every row has a download button.

• Visual — Characters (+ opens Add a character), Product / Service (+ uploads a photo right away), Assets (+ opens Add an asset), Locations (+ opens Add a location), and Storyboards once there are any. A footer line reads <n> of 6 photos — the product, the assets and the locations share them.
• Rows carry origin chips (Your photo, Library character, Generated, Not generated yet) and scene chips (S1, S2, +N, or No scene). Empty groups explain themselves, for example No product photo — the model invents the product. Upload one to show yours.
• Audio — Spoken lines: one row per speaking scene with its length, Reading…, or Not read yet, an inline player, and a download (scene_01_line.wav).
• Video — Scene clips, Finished cuts, and Final video.
• Data — the plan file, Ad plan.

Settings

The ad's render settings. Defaults: Lite, 720p, captions off.

• Render — Quality Lite (Seedance 2.0 fast · 720p), Pro (Seedance 2.0 · 720p or 1080p) or Max (Seedance 2.5 · 720p or 1080p), each with its price per 10 seconds. On a crazy hook: Max renders the chaos with more believable physics.
• Format — Aspect ratio (read-only, Set by the plan) and Resolution 720p / 1080p (not on Lite: 1080p is available on Pro and Max; HEVC, larger files). Switching to Lite drops back to 720p.
• Captions — the same caption controls as a story, with the note Captions follow each spoken line. Texts you add to a scene are burned in whether captions are on or off. Turning captions on or off, or changing their style, deletes the finished cuts and the final video; the finish rebuilds them without re-rendering any clip.
• Voice — Speech pace (Slow to Fast). Unlike the rest of the tab it applies at once, after a confirmation, and re-reads every line (see "Speech Pace").

The render and caption changes are staged until Apply changes. While they are unsaved the footer reads Unsaved render settings and rendering is blocked (Save the render settings first). Apply is disabled during a run (The video is being generated. Apply these changes once it finishes.).

Footer

• During a run: Stop Generation.
• While clips are in the cloud: a sky chip, Aificient Cloud: rendering 2/4 or Aificient Cloud: 3 in queue, opens the render queue (Captions, texts and the final cut run automatically once every clip is delivered.).
• Otherwise: Resume Generation, amber when something is missing. It becomes Stitch Final Video when every clip exists and only the finish is owed (after a text, transition, or caption change, or a deleted final video) and then runs directly.

The Resume Generation menu, Generate video on, has a single runtime, Aificient Cloud (Seedance), with the tier, resolution, and estimate, and two actions:

When something is missing the primary action explains it: … Generate the assets first (Generate storyboard only). or … Open the scene, check its prompt and save it. While scenes are already in the cloud queue, the button can still render the others: Renders the N scenes nothing renders yet; the ones in the cloud queue keep rendering there.

Generating and Resuming a UGC Ad

A run goes through these steps and keeps everything that already exists; only the gaps are filled.

1. Place photos — every place with a description but no photo gets one (Adding the missing place photos).
2. References — every image not generated yet (portraits, product shots, assets, places), three at a time (Generating references · x of y).
3. Lines — the voice reads the lines of the scenes this run renders that have no take yet (Reading the lines).
4. Storyboards — a full run (Start generation, Generate video) sketches every scene that has neither a sheet nor a clip, three at a time; a failed sheet is reported and the run goes on. Generate storyboard only stops here.
5. Check — nothing renders while an image is missing, a scene needs review, or a prompt names an image it does not send: No video until every asset is ready. <what is missing>.
6. Render — each scene is submitted to Aificient Cloud and charged when it is accepted; its old clip, finished cut, and the final video are removed at that moment. The run ends at submission; the clips keep rendering in the cloud, even if you close the app.
7. Finish — once every clip is delivered, automatically: the finished cuts (captions are timed from the clip itself; texts are burned in, locally and free on the desktop, on the server on the web), then the stitch (local and free on the desktop; on the web on Aificient Cloud for 0.1 credits). Hard cuts are the default, with each scene's own transition where you set one, and any trims you saved cut the clips at this point. At a hard cut the sound of both clips is faded over a few hundredths of a second, so there are no clicks; across a transition the two soundtracks crossfade for its length. A failed burn falls back to the plain clip (Scene N: <error> — its texts are not in the final video).

Reopening a project whose clips all arrived while the app was closed runs the owed finish once by itself; it never renders a scene on its own. Stop Generation stops the local steps; scenes already in the cloud keep rendering and are paid. A queued one can still be cancelled (and refunded) from the render queue.

Desktop notifications: UGC ad prepared (<name>: the storyboards are ready to review.) and UGC ad ready (<name>: the final video is ready.), or Crazy Hook ….

Cost: The plan card, the settings screen, the menu, and every render button show the estimate before you spend anything. Running out of credits opens the out-of-credits card.

• Clips — per second at the saved tier: Lite is the cheapest, then Pro, then Max; 1080p costs more.
• Storyboards — per sheet.
• Generated images and line readings — each charged separately.
• The stitch — free on the desktop; 0.1 credits on Aificient Cloud on the web.

Render Messages

What UGC Ads Do Not Have

• Your own GPUs. Clips render on Aificient Cloud only; there is no local or rented runtime choice.
• A visual-style preset, scene images, or a narrator mix. The look comes from your photos and the creators; the voice is read per line and lip-synced.
• Publishing. There is no Publish button; download the final video from the Output node, the full-screen viewer, or the Assets tab.
• The Generating on <place> chip. A UGC ad's sidebar does not show another device's run.

UGC Limits