Boreal-H3 vs Boreal: Which of Creatify's Two Ad Video Models You Actually Want

The short answer is that Boreal-H3 and Boreal are not competing products at different quality levels — they are two tiers of the same idea, and the split is not resolution or realism. Boreal, launched September 15, 2026, is built for volume: a cent a second, 720p-class, text-to-video and image-to-video, tuned so a twenty-shot storyboard costs about a dollar. Boreal-H3, live September 30, 2026, is built for the shots a campaign hangs on: four cents a second at launch, up to 2K, and a reference workflow that takes nine images, three video clips and three audio clips into a single shot. Both are post-trained on MiniMax H3 through the same Creatify pipeline. The question is not which is better. It is which class of shot you are buying for.

If you only read one paragraph

Boreal wins on cost and on the formats where a talking head or a simple product pan is the whole job. Boreal-H3 wins on anything where the product has to stay exactly itself, the creator has to stay the same person across shots, the camera has to be directed, or the shot needs references — and you pay four times the rate, at launch, for that control. If your pipeline generates variants at volume and picks the best, Boreal is the cheaper loop because the loop runs more times than the output. If your pipeline produces a handful of campaign shots that have to be right on the first take, Boreal-H3 is the one whose failure modes cost more than its rate.

The two, dimension by dimension

• Price — Boreal $0.01/s vs Boreal-H3 $0.04/s at 768P, its launch rate, which Creatify states is 50% off MiniMax H3's $0.08/s list.

• Higher tiers — Boreal 720p-class only vs Boreal-H3 $0.12/s at 1088P and $0.26/s at 2K.

• Inputs — Boreal text-to-video and image-to-video vs Boreal-H3 text-to-video, image-to-video with an optional end frame, and reference-to-video.

• References — Boreal none vs Boreal-H3 up to 9 images, 3 video clips and 3 audio clips in one shot.

• Speed — Boreal quoted 1:1 with realtime vs Boreal-H3 measured at 1.10 generation-seconds per video-second at 768P, so roughly 10% over realtime at the tier both share.

• Underlying base — Boreal its own post-trained base vs Boreal-H3 MiniMax H3, the general omni-modal model both are descended from.

• Evidence — Boreal a 40-case blind test against its own base, an 81% preference overall and 60% on single-person talking clips vs Boreal-H3 an internal production-brief test against MiniMax H3, brief completion 27.8% to 50.0% and identity match 83.3% to 94.4%.

Read the last two lines together and the tiers stop looking like a ladder. The two models' published evidence sets are different shapes — Boreal's is a preference test, Boreal-H3's is a completion-rate test — so the natural instinct to compare their headline percentages is the wrong instinct. They are not the same measurement.

Where the extra three cents a second goes

Boreal-H3's price is four times Boreal's, and the page is specific about what that buys. The reference workflow is the structural difference: Boreal takes a prompt or an opening frame, and everything else about the shot is inference. Boreal-H3 takes up to nine images, three video clips and three audio clips as conditioning, which is what allows a consistent product, a consistent creator and a directed camera across a shot rather than a single lucky clip.

The quality numbers are stated against the base both models share. On difficult production briefs, Creatify reports subject consistency — identity match — moving from 83.3% to 94.4%, and visible defects falling from 1.11 to 0.33 per clip, a drop of about 70%. Against six frontier video models on a normalised ads-quality index where MiniMax H3 is set to 100, Boreal-H3 is reported at 133.3 with 85.3% reference fidelity. Those are vendor-run figures on vendor briefs. They are not an independent bake-off, and the anchor at 100 is a model the same company post-trained.

The most direct evidence of what the price buys is a fifteen-second image-to-video clip of a squishy toy: 84 seconds on Boreal-H3 against 411 seconds on Seedance 2.5, run through that competitor's public API. That is a vendor measurement of a vendor pipeline against a competitor's endpoint, and it will not settle the question on its own — but it is the right shape of claim to check, because render time is the one comparison a competitor's public API cannot misreport.

Where Boreal is still the better buy

Three places, and they are not consolation prizes. The first is volume: at a cent a second, a generative loop that produces twenty variants to find one good one costs less on Boreal than the same loop costs on Boreal-H3's cheapest tier. For pipelines where the output is the winner of many attempts rather than a single directed shot, that arithmetic does not change with quality.

The second is talking heads. Boreal's own launch reported its weakest preference precisely there — 60% on single-person talking clips against 83% on product ads and 85% on creator and UGC scenes — and a presenter reading a script is the single most commoditized format in the ad stack. If that is the whole job, the cheaper model was already good enough, and Boreal-H3's reference machinery is being paid for and not used.

The third is resolution honesty in the other direction. Boreal is 720p-class and its launch said so plainly, measuring itself against a 1080p flagship only to concede where it loses. Boreal-H3 starts at 768P and goes to 1088P and 2K. If your delivery is not resolution-sensitive, that ladder is one you are paying for without climbing.

The floor both models share

Here is the part a straight comparison hides. The improvement Boreal-H3 reports is measured against a floor the same company set one release earlier: the 83.3% identity start line it has to beat is the exact figure Boreal's launch used a fortnight before, on the same base lineage. So the gain is real and relative to what Creatify was serving, and the anchor underneath it is still MiniMax H3 — a model from a different lab entirely, and one we do route. Neither Boreal nor Boreal-H3 has a published number from anyone who does not sell it. Compare the two on price, inputs and reference budget and you are comparing facts. Compare them on the quality percentages and you are comparing one vendor's method against the same vendor's method.

The router question

Neither Boreal nor Boreal-H3 is in our catalogue, and this is not a piece that will pretend otherwise. OrcaRouter does not serve video generation models. Where the pair matters to a router is the same place it mattered for Boreal's launch: an ad pipeline is one video call wrapped in a lot of text work. Brief expansion, the hook and script variants, the compliance pass, the caption set, and the logic that decides which cut runs where — that half scales with the number of variants, not the number of clips, and it is the half we serve. Those calls run at provider list price with 0% markup on one key, with automatic failover across providers when a single endpoint degrades.

If you are choosing between these two tiers for the video itself, the choice is the one above: volume and cost against control and references. If you are building the loop around them, the routing layer is the part you can hold constant across whichever tier wins.

What would change this answer

Three things, and none of them is a bigger vendor demo. First, an independent run. Boreal-H3's evidence is an internal completion-rate test; Boreal's is an internal preference test; both are self-reported, and until a third party generates the same briefs on both models, the quality half of this comparison is provisional.

Second, the price. Boreal-H3's $0.04 is a launch offer defined as half of MiniMax H3's list. If it holds at list instead, Boreal-H3 sits at eight times Boreal rather than four, and the volume case for Boreal gets stronger still — the tier gap is the one number here that Creatify has not committed to past the launch window.

Third, whether the tiers converge. Boreal's own roadmap pointed at reference-conditioned generation for multi-shot work, which is exactly the capability that separates the two models today. If that arrives on the cheap tier, the four-times premium is about resolution and defect rate and not much else.

Until one of those moves, the honest split stands: buy Boreal for the loop, buy Boreal-H3 for the shot, and treat every quality figure from either of them as the vendor describing its own work.