- Learn
- Formats & Exhibition
- AI Video Ad
Formats & Exhibition · AI advertising
AI Video Ad
External-tool reference examples from the source experiments, not a claim of current FlashBoards model availability or installed automation. Use your own supported tools and authorized media; settings, timings and comparisons describe the tested runs.
Also called: AI commercial, AI UGC ad, AI product video, AI packshot, AI ad creatives, AI banner
An ad has one job: show the product, the same in every frame. Below, a can of cold brew we made up — OSKA — in three formats from one product sheet: Mara’s story to her followers, a 15-second commercial on two engines side by side, and a post, a story and a banner — with every prompt and time.
Video script
Hi, I'm Yaro. Today: an AI video ad for a drink that doesn't exist. OSKA, sparkling cold brew. Fake can, real ads.
An ad has one job: the same product in every frame. Describe the can in words, and the label drifts, and the can grows a twin.
So step one: draw the product once. Four sides, a label close-up, the can in cold droplets. That sheet goes into every step.
Step two: UGC. Mara records her voice first, in a whisper, like a confession. Then one MiniMax H3 Max request: her sheet, the can, a start frame, the voice. Fourteen seconds.
Pro tip: never mention the phone. Say the camera is never seen. Very ninja.
Step three: a fifteen-second commercial. The end card with the slogan is drawn as a picture. Text belongs to pictures. The video just holds it.
Same spot, two engines. H3 Max: nine seconds, with a reported cut approximately two seconds early. Seedance 2.5: seven minutes, reported cuts approximately one second early and half a second late, and its own hiss and splash. Compare the draft and quality routes.
Step four: a post, a story, a banner. GPT Image 2.5 at quality medium copies every line in quotes word for word. So proofread. Typos are forever.
The three formats share the same product reference.
Every prompt is right below. Scroll down. I'll be here, trying to drink a pixel can.
- What it does
- A UGC video, a commercial and static creatives for one product, the product identical everywhere.
- Use
- GPT Image 2.5 for the product sheet and creatives, MiniMax H3 Max for UGC and drafts, Seedance 2.5 for the final spot, ElevenLabs for voice and sound.
- Keep
- one product sheet as a picture in every step.
One product sheet, three formats
OSKA is a can of sparkling cold brew we made up for this test. The same picture of the can made a creator’s story, a 15-second commercial and three creatives — and the label never changed.
- Product
- one product sheet — four sides, the label, the hero state — fed as a picture into every step
- UGC
- MiniMax H3 Max 9:16 · Mara’s voice on ElevenLabs v4 · 14 s
- Commercial
- Seedance 2.5 720p for the final · H3 Max 480P for drafts
- Statics
- GPT Image 2.5 medium · the offer text drawn in one pass
ad — a reference guide linking the steps
For a product and character, this external reference guide covers how to draw the product sheet, write and voice the UGC script, build the commercial from keyframes and music, and draw the posts, stories and banners — with the exact settings and prompts from this page.
⬇Download reference guideadA reference guide connecting the creative steps — sheet · UGC · spot · statics- 1ad-product-sheetOne picture of the product, every side
- 2ad-ugcThe creator talks, the product holds
- 3ad-spotSix shots on the beat, H3 or Seedance
- 4ad-staticFeed, story, banner with the offer
The short answer
An AI ad lives or dies on the product looking the same in every frame. Draw the product once as a sheet and give that picture to every step. Then compare formats: a creator’s story on MiniMax H3 Max, a commercial drafted on H3 Max and finished on Seedance 2.5, and banners whose text the image model draws itself.
- Product✕ the can described in words✓ one sheet with every side, as a picture
- UGC✕ “holding her phone”✓ “filmed at arm’s length, the camera never seen”
- Commercial✕ the slogan asked of the video✓ the end card drawn as a keyframe
- Final spot✕ H3 cuts up to 1.5 s off the beat✓ Seedance 2.5 within 0.25 s, its own sound effects
- Statics✕ quality low for text✓ quality medium, every line in quotes
Draft or quality: Seedance 2.5
The external method on this page uses MiniMax H3 for drafts and many takes. Seedance 2.5 reference-to-video accepts the same pictures and sound as references (@Image1, @Audio1), up to 30 seconds a call, with its own synced sound at 480p, 720p or 1080p. The source preferred Seedance for quality. The product-commercial comparison below reports cuts approximately one second early and half a second late, finer liquids and detail, and its own sound effects in sync. This side-by-side result is from a product commercial, not a measurement on every format.
1 · The product as a picture
GPT Image 2.5 Flare · text-to-image · mediumModels redraw a product described in words: the label drifts, the colours shift, the can grows a second one. So the product is drawn once — four sides, a big close-up of the label, the can covered in cold droplets — and that sheet goes into every later image and video as picture 1. Every word of the label is written in quotes in the prompt.
Product reference sheet of one fictional drink can, on a plain light grey studio background, soft even light, no other objects, no people. The product: OSKA sparkling cold brew coffee, a slim 330 ml aluminium can with a matte black finish; the word "OSKA" in big bold white sans-serif capitals running vertically up the front; one thin bright orange horizontal band around the can a third of the way from the bottom; above the band, small white text "SPARKLING COLD BREW"; below it, small white text "330 ML"; a plain silver top with a ring pull. Layout: four views of the same can in a row — front, right side, back (small white text "Cold brew. Zero chill." and a few lines of fine print), left side; below them, on the right, a large close-up of the front label, and on the left the same can covered in cold condensation droplets. Photoreal packshot, exact same can in every view, crisp text, no extra logos.
2 · UGC with your character
ElevenLabs v4 · GPT Image 2.5 · MiniMax H3 Max 9:16Mara, our recurring character, tells her followers about OSKA like a friend would: a whispered confession, the can cracked open at the lens, a sip, the benefit, a wink and “link’s in my bio”.
- The voice first. Hope on ElevenLabs v4 with tags —
[whispers],[excited],[playful]— 13.8 s; Scribe gives the word times. - A start frame with GPT Image 2.5: her sheet and the product sheet as pictures, a sunny kitchen, the can on the counter label-out, “as if she films herself at arm’s length, the camera itself never seen”.
- One H3 Max request: Image 1 her sheet, Image 2 the product sheet, Image 3 the start frame, Audio 1 her voice; four beats on the word times, two jump cuts.
- Sound: the voice with a short room echo, kitchen room tone 13 dB under it and the can-crack sound effect where she opens it.
The can never changed — the label, the orange band, “330 ML” — and her voice landed with a shift of 0.00 s.
[whispers] Okay, I need to confess something... [normal] I quit my three coffees a day. For this. [excited] Cold brew, but sparkling. No sugar. And it actually wakes me up. [playful] Link's in my bio. Don't tell my barista.
UGC · the full H3 Max prompt
Source creative prompt notation below, not a JSON API request.
{
output: a vertical 9:16 UGC ad, 14 seconds, as if Mara films her own story at arm's length — the camera itself is never seen; one continuous take with the natural shake of a hand-held camera, two quick jump cuts at the given times; Audio 1 is exactly her voice and plays from 00:00.000,
references: {
Image 1: only Mara's face, hair and clothes,
Image 2: only the drink can — exactly this can, its OSKA label and text, every time it is on screen,
Image 3: the first frame — the kitchen, the light and Mara as the video starts,
Audio 1: Mara's voice; she says every word, lips in sync
},
character: { Mara: lively, charming, a little mischievous; her head, eyes and face move all the time, natural micro-expressions, she talks to the viewer like to a friend },
environment: { a bright lived-in kitchen on a sunny morning, soft daylight from the window, exactly as on Image 3; sound: only Audio 1 },
style: real creator footage, natural skin, real light, no text on screen, no logos except the can's own label,
shots: [
{ from: 00:00.000, to: 00:02.800, action: she leans toward the camera and lowers her voice, a conspiratorial half-smile, eyebrows up, saying "Okay, I need to confess something." },
{ from: 00:02.800, to: 00:06.000, action: she grabs the OSKA can from the counter, cracks it open with a sharp hiss right at the camera, and says "I quit my three coffees a day. For this.", lifting the can toward the lens on "this" },
{ cut_at: 00:06.000, from: 00:06.000, to: 00:11.000, action: a jump cut closer: she takes a sip, eyes widen, then turns the cold, wet can slowly next to her face so the OSKA label faces the camera, saying "Cold brew, but sparkling. No sugar. And it actually wakes me up." },
{ cut_at: 00:11.000, from: 00:11.000, to: 00:14.000, action: a jump cut: she holds the can against her cheek, winks, and says "Link's in my bio. Don't tell my barista." with a cheeky grin, then laughs }
],
rules: [the can is always exactly the can of Image 2; only one can; Mara is the only person; the camera is never seen; no phones; no subtitles]
}
3 · The 15-second commercial
keyframes · ElevenLabs Music · H3 Max and Seedance 2.5A product film with no people but one hand: a droplet over the label, the can cracking open, cold brew pouring over ice, an orbit on the bass drop, a splash into ice water, the end card. Three keyframes set the look — and the end card, with its slogan, is drawn as a picture, so the video only has to hold it.
Music from ElevenLabs Music: 15 s of dark minimal electronic at 120 BPM with one big drop at 8 s — the cuts are written on its bars.
The same commercial on two engines
| MiniMax H3 Max reference-to-video | Seedance 2.5 reference-to-video | |
|---|---|---|
| Inputs | Image 1 product sheet, Image 2–4 keyframes, Audio 1 music — a JSON prompt | @Image1 product sheet, @Image2–4 keyframes, @Audio1 music — plain text, a paragraph per shot |
| Resolution | 832 × 480 | 1280 × 720 |
| Cuts (planned 2.5 · 4.5 · 8.0 · 10.0 · 12.0) | 2.4 · 4.0 · 6.0 · 10.7 · 12.0 | 2.4 · 4.75 · ~7 · 10.0 · 12.5 |
| The can and the end-card text | exact | exact |
| Sound | plays our music as given — we add the effects | re-composes the music and adds its own hiss, pour and splash in sync |
| Time | 9 s | 428 s |
| Use for | drafts, quick versions | the final spot |
H3 Max · the full prompt
Source creative prompt notation below, not a JSON API request.
{
output: a premium 16:9 TV commercial for a drink, 15 seconds, six shots with hard cuts exactly at the given times, on the beat of Audio 1; no people except one hand,
references: {
Image 1: only the can — exactly this can, its OSKA label, orange band and text, every time it is on screen,
Image 2: shot 1 starts exactly like it,
Image 3: shot 3 starts exactly like it,
Image 4: shot 6 is exactly this frame, with exactly this text,
Audio 1: the music; every cut and every hit lands on its beat
},
environment: { a dark product studio, black wet glass, deep black background, a hard orange rim light, cold condensation everywhere },
style: high-end commercial cinematography, macro lenses, crisp detail, real liquids and ice, smooth precise camera moves, no text except in shot 6,
shots: [
{ shot: 1, from: 00:00.000, to: 00:02.520, image: Image 2, camera: an extreme macro, a very slow push-in along the can, action: a cold droplet slides down over the OSKA lettering and catches the orange rim light },
{ shot: 2, cut_at: 00:02.520, from: 00:02.520, to: 00:04.520, camera: an extreme close-up of the top of the can, static, action: a finger pulls the ring — the can cracks open with a burst of fine mist and fizz right on the beat },
{ shot: 3, cut_at: 00:04.520, from: 00:04.520, to: 00:08.020, image: Image 3, camera: a slow dolly-in, slight slow motion, action: dark cold brew pours from the tilted can into the tall glass of ice, bubbles rising, a tan foam building, droplets running down the glass },
{ shot: 4, cut_at: 00:08.020, from: 00:08.020, to: 00:10.020, camera: a fast orbit around the standing can, action: on the bass drop the orange rim light flares and sweeps across the wet can as the camera circles it, the OSKA label flashing past },
{ shot: 5, cut_at: 00:10.020, from: 00:10.020, to: 00:12.020, camera: a low angle at the surface of icy water, slow motion, action: the can plunges into a tub of ice water, a crown splash and ice cubes bursting outward, the label rising back into view },
{ shot: 6, cut_at: 00:12.020, from: 00:12.020, to: 00:15.000, image: Image 4, camera: static, a barely noticeable push-in, action: the end card exactly as Image 4 — the can centred among ice cubes, the text "OSKA" and "Cold brew. Zero chill." beside it, droplets glistening; it holds to the end }
],
rules: [the can is always exactly the can of Image 1; only one can; the text appears only in shot 6 and exactly as in Image 4; no other logos; no split screen]
}
Seedance 2.5 · the full prompt
A premium 16:9 TV commercial for the drink can @Image1 — exactly this can, its OSKA label, orange band and text, every time it is on screen; only one can. Six shots with hard cuts on the beat of the music @Audio1. A dark product studio, black wet glass, deep black background, a hard orange rim light, cold condensation everywhere. High-end commercial cinematography, macro lenses, real liquids and ice, smooth precise camera moves. No people except one hand. No text except in the last shot. Shot 1 (0–2.5 s), starts exactly like @Image2: an extreme macro, a very slow push-in along the can; a cold droplet slides down over the OSKA lettering and catches the orange rim light. Shot 2 (2.5–4.5 s): an extreme close-up of the top of the can; a finger pulls the ring and the can cracks open with a burst of fine mist and fizz right on the beat. Shot 3 (4.5–8 s), starts exactly like @Image3: a slow dolly-in, slight slow motion; dark cold brew pours from the tilted can into the tall glass of ice, bubbles rising, a tan foam building. Shot 4 (8–10 s): on the bass drop, a fast orbit around the standing can; the orange rim light flares and sweeps across the wet can, the OSKA label flashing past. Shot 5 (10–12 s): a low angle at the surface of icy water, slow motion; the can plunges into a tub of ice water, a crown splash, ice cubes bursting outward. Shot 6 (12–15 s): the end card exactly as @Image4 — the can centred among ice cubes, the text "OSKA" and "Cold brew. Zero chill." beside it, droplets glistening; it holds to the end. Sound: the music @Audio1 with crisp sound effects — the can hiss, the pour over ice, the splash.
A 15-second instrumental music bed for a premium energy-drink TV commercial: dark, cool minimal electronic, 120 BPM, a tight kick and crisp hi-hats, a deep sub bass, a short riser in the middle and one big bass drop hit on the last bar, then a clean ending. No vocals.
4 · Static creatives
GPT Image 2.5 Flare edit · quality mediumPosts, stories and banners in one pass each: the product sheet as picture 1 (and Mara’s sheet for the story), the scene, and every line of text in quotes with its size, colour and place. At quality medium all three came out with the text exactly as written — so write the offer the way it should read.
Picture 1 is only the look of the drink can: take the can exactly as it is, with its OSKA label, orange band and text. A square social media ad, 1:1: the can from picture 1, covered in cold droplets, floating slightly tilted in the centre against a flat, bold orange background that matches the orange band of the can, a soft shadow under it, a few ice cubes and splashes of dark cold brew frozen in mid-air around it. Big clean black headline at the top: "Zero chill." Small black line at the bottom: "OSKA — sparkling cold brew". Modern minimal graphic design, crisp typography, perfectly sharp text, no other text or logos.
Picture 1 is only the look of Mara: her face, hair and clothes. Picture 2 is only the look of the drink can: take the can exactly as it is, with its OSKA label, orange band and text. A vertical 9:16 Instagram story ad: Mara at a sunny window early in the morning, holding the cold can from picture 2 next to her face with a sleepy, delighted half-smile, warm golden light, real skin, shallow depth of field. At the top, big white bold text with a soft shadow: "My 6 AM secret". At the bottom, a rounded orange button with black text "Try OSKA". Clean social design, perfectly sharp text, no other text or logos.
Settings and measurements
fal · ElevenLabs| Step | Tool | Time |
|---|---|---|
| Product sheet | GPT Image 2.5 Flare text-to-image, medium, 2048 × 1152 | ~30 s |
| UGC voice | ElevenLabs v4 · Hope | 5 s |
| UGC start frame | GPT Image 2.5 Flare edit, low, 1152 × 2048 | 27 s |
| UGC video, 14 s | MiniMax H3 Max reference-to-video, 480P, 9:16 | 8 s |
| Three keyframes | GPT Image 2.5 Flare edit, medium, 1920 × 1088 | 22 s each |
| Music and sound effects | ElevenLabs Music, Sound Effects | seconds |
| Commercial, 15 s · draft | MiniMax H3 Max, 480P | 9 s |
| Commercial, 15 s · final | Seedance 2.5, 720p | 428 s |
| Three creatives | GPT Image 2.5 Flare edit, medium | 22–32 s |
What works best
our conclusionsOne product sheet, three formats
OSKA is a can of sparkling cold brew we made up for this test. The same picture of the can made a creator’s story, a 15-second commercial and three creatives — and the label never changed.
- Product
- one product sheet — four sides, the label, the hero state — fed as a picture into every step
- UGC
- MiniMax H3 Max 9:16 · Mara’s voice on ElevenLabs v4 · 14 s
- Commercial
- Seedance 2.5 720p for the final · H3 Max 480P for drafts
- Statics
- GPT Image 2.5 medium · the offer text drawn in one pass
- Draw the product once, then only show it. One sheet with every side keeps the label exact in every image and video.
- Text belongs to pictures. The end card and the banners are drawn by the image model; the video only holds them.
- UGC is a voice first. Record the line, take its word times, and tie every action to a word.
- Never mention the phone — say the camera is never seen.
- Draft on H3 Max, finish on Seedance 2.5 — 720p, closer cuts and sound effects in sync.
- Quality medium for any text, and write the offer exactly as it should read — the model copies it word for word.
The product sheet, the UGC, the commercial and every creative on one board.
FAQ
5 questionsHow do I keep my product the same in an AI video ad?
Make one product sheet — every side, a close-up of the label, the product in its hero state — and give it as a picture to every image and video model, with “exactly this product, its label and text, every time it is on screen; only one”. Words alone let the label drift.
How do I make a UGC ad with an AI character?
Voice the script first (ElevenLabs v4 with emotion tags), make a start frame of the character with the product, then generate on MiniMax H3 Max reference-to-video in 9:16 with the character sheet, the product sheet, the start frame and the voice, each action tied to the time of its words. Add room tone and the product’s sound.
Seedance 2.5 or MiniMax H3 for a product commercial?
H3 Max is the draft: 9 seconds for 15 s at 480P, with a reported cut approximately two seconds early. Seedance 2.5 is the source’s final: about 7 minutes for 15 s at 720p, reported cuts approximately one second early and half a second late, finer liquids and its own sound effects in sync.
Can an image model write the ad text?
Yes. GPT Image 2.5 at quality medium drew every headline, button and offer exactly as written in quotes in the prompt. Write the text grammatically — the model copies it word for word.
What did the ad experiment include?
The test used one product sheet, a 14 s UGC video, a 15 s commercial on H3 Max and three static creatives; the final commercial was also generated with Seedance 2.5 at 720p.
Related
3- AI Reels & ShortsFormats & Exhibition
A recurring AI character that talks to camera: voices, cloning, lip-sync on H3, room sound.
- AI Music VideoFormats & Exhibition
A three-act music video on frames from one film, every face replaced, MiniMax H3 Max with lip-sync.
- AI Film Scene RemakeFormats & Exhibition
A film scene remade with your characters: the film’s edit, the actors’ timing in your voices.