NEW

Boreal-H3

The best video generation model built for ads. Precise control over how a shot starts, where it lands, and which creative references define it, while staying fast and cost-efficient enough to keep iterating.

TRY IN CREATIFY

READ THE TECH REPORT

BRIEF SUCCESS

pts

28% → 50% on controlled shots

IDENTITY MATCH

up from 83% in the same evaluation

END-TO-END QUALITY

pts

43.7 → 69.0 acceptability

ITERATION EFFICIENCY

−

generation time and estimated cost

01 — THE BOREAL UPGRADE

From scalable generation to directed generation

Boreal proved that advertising video could be fast, affordable, and good enough to generate at scale. Boreal-H3 takes the next step: a substantial jump in quality for the shots that demand tighter control, stronger identity, and more faithful use of product and scene references.

Boreal made more generations possible. Boreal-H3 makes more of those generations usable.

BOREAL

Maximum speed and value

Best for short, straightforward shots and high-volume exploration when turnaround and cost come first.

BOREAL-H3

Higher quality and deeper control

Best for complex briefs, product-critical shots, creator consistency, directed transitions, and multi-reference work.

The product choice is simple: use Boreal for the fastest, lowest-cost exploration. Use Boreal-H3 when the shot needs more precision. Both are built for advertising teams that create variants, not one-off demos.

02 — TWO CONTROLS

Two ways to control the shot

Boreal-H3 is available through two separate entries in Model Playground. Pick the workflow that matches what you need to control: the shot’s endpoints, or the creative references inside it.

OPENING FRAME

→

BOREAL-H3 RESULT

KEYFRAME CONTROL · IMAGE TO VIDEO

Keep the person and composition from frame one.

Start from an approved portrait. Boreal-H3 preserves the subject and visual style while adding natural motion and a spoken performance.

Land the shot.

Add an ending frame when the shot must land on a precise destination.

Direct any shot type.

Talking-head, lifestyle, product, and transition shots.

Stay on-model.

Identity and framing stay consistent from the approved visual into the clip.

REFERENCE 1 · FRONT

REFERENCE 2 · BACK

→

BOREAL-H3 RESULT

MULTI-REFERENCE CONTROL · REFERENCE TO VIDEO

Turn multiple product views into one coherent shot.

Here, separate front and back references become one complete 360° product video, preserving the pattern, striped reverse, trim, and tassels.

Combine every ingredient.

Product, person, styling, scene, and voice references in one shot.

Show what one image can’t.

Product details that no single image can capture.

Keep it all consistent.

Multiple creative ingredients stay consistent through the shot.

03 — SIX REASONS

Six reasons

to pick Boreal-H3

What the model does well, in the order that matters most for advertising. Each point is backed by first-round tests from Creatify’s video design team, run on real customer products inside AdFlow.

Your product stays your product.

Label text, shape and details stay accurate from start to finish. One reference image is enough, and the model picks the reference angle that shows only accurate details.

AI GENERATED

SINGLE REFERENCE IMAGE

One reference was enough to keep the product consistent through the whole clip.

AI GENERATED

UGC TALKING HEAD WITH PRODUCT

Product accuracy and dialogue delivered cleanly in the same take.

Tuned for ads, not general video.

Trained specifically for product consistency and UGC, so creator-style ads and testimonials come out usable on the first take.

AI GENERATED

UGC AD · FIRST TAKE

A creator-style ad straight from the model on the first try, not cherry-picked.

AI GENERATED

TESTIMONIAL AD · FIRST TAKE

Straight from the model on the first try, not cherry-picked.

Built for UGC ads.

Talking heads deliver complex lines without stumbling, and two-person conversations work too. It even picks a fitting accent from context.

AI GENERATED

TALKING HEAD · COMPLEX LINES

Clear speech with no stumbling, even on lines that are hard to deliver.

AI GENERATED

TWO-AVATAR CONVERSATION

Dialogue lands correctly, with an accent chosen from the reference and context.

Fast and cheap, so you can iterate.

Generation is faster than most tools and much cheaper, so you get to a winning cut in fewer rounds.

AI GENERATED

DELTA AD · ENGLISH TO GERMAN

Came back faster than most tools and at a much lower cost.

AI GENERATED

DELTA AD · SECOND CUT

A second cut from the same localization task.

Directable camera.

It follows the angles, moves and scene order you ask for, with natural transitions and no loss of speech clarity.

AI GENERATED

DYNAMIC CAMERA · MULTI-ACTION

Follows the requested angles and shot progression while speech stays clear.

AI GENERATED

MORE DYNAMIC PROMPT

The same setup pushed with a more dynamic camera and action prompt still holds up.

Physics that sells.

Actions have real results: the vacuum actually leaves the floor clean. Liquid, product handling and transparent packaging look believable.

AI GENERATED

VACUUM · CAUSE AND EFFECT

The surface is visibly clean after the pass.

AI GENERATED

GUMMIES · TRANSPARENT PACK

Transparent packaging rendered correctly.

04 — HEAD TO HEAD

Same brief, same inputs, head to head

An ad can look beautiful and still be unusable: the bottle changes shape, the label turns into gibberish, the creator becomes someone else, or the product never does what the brief asked. Boreal-H3 is refined on real advertising briefs to avoid exactly those failures. These pairs put it against Seedance 2.5 and Kling 3.0 on identical briefs and reference images.

SAME BRIEF · SAME REFERENCES · BOREAL-H3 ON THE LEFT

WATCH FOR 01

Does the product keep its shape?

The bottle stretches, the pack gains a flap, the pillow loses its tassels. Boreal-H3 keeps geometry and details true to the reference.

WATCH FOR 02

Is the label still readable?

Printed text is the first thing a brand checks. Boreal-H3 holds label text and logos legible and stable through the camera move.

WATCH FOR 03

Is it still the same person?

A face that drifts between frames is a clip nobody can post. Identity match rose from 83% to 94% in the same evaluation.

WATCH FOR 04

Did the action actually happen?

The brief said open the can, turn the product, land on the hook. Brief success on controlled shots rose from 28% to 50%.

Boreal-H3

Seedance 2.5

01 / 05Squeezing a squishy toyImage-to-video, 15s. The beads have to burst into liquid inside the translucent body. Render time: Boreal-H3 84s, Seedance 2.5 411s.

05 — PRICING

Simple, per-second pricing.

Pay only for the seconds you generate. At 768P, Boreal-H3 costs half the MiniMax-H3 list price.

Per output second768P1088P2K

Text to videoPrompt only · 6 aspect ratios

~~$0.08/s~~ MiniMax-H3$0.04/s0.4 credits/s

$0.12/s1.2 credits/s

—

Image to videoFirst frame, optional last frame

~~$0.08/s~~ MiniMax-H3$0.04/s0.4 credits/s

$0.12/s1.2 credits/s

$0.26/s2.6 credits/s

Reference to videoUp to 9 images, 3 videos and 3 audio clips

~~$0.08/s~~ MiniMax-H3$0.04/s0.4 credits/s

$0.12/s1.2 credits/s

—

BOREAL-H3

The best video generation model built for ads.

Control how a shot starts and where it lands, combine product and creator references, and iterate fast enough to find the winning cut.

TRY IN PLAYGROUND

BOREAL-H3

The best video generation model built for ads.

Control how a shot starts and where it lands, combine product and creator references, and iterate fast enough to find the winning cut.

TRY IN PLAYGROUND

BOREAL-H3

The best video generation model built for ads.

Control how a shot starts and where it lands, combine product and creator references, and iterate fast enough to find the winning cut.

TRY IN PLAYGROUND

06 — ABOUT THE COMPARISONS

Seedance 2.5 and Kling 3.0 were run through their public APIs with the identical brief text and reference images. Boreal-H3 is shown from Creatify’s served renders at 1088P. No clip was edited. Original files are available on request.

Render time is the wall-clock seconds to generate each clip from the same brief.

The strengths in section 3 are backed by first-round tests from Creatify’s video design team (Sep 24, 2026), run on real customer products inside AdFlow.

Headline figures (brief success, identity match, end-to-end quality and iteration efficiency) come from Creatify’s internal evaluation of Boreal-H3.

Creatify · Boreal-H3. Updated September 29, 2026.

The research lab building lifelike AI video models.