Creatify's Boreal Costs a Cent a Second, and the Launch Post Says Where That Breaks

Creatify launched Boreal on September 15, 2026, and led with a price: one cent per second of finished video, or about five cents for a five-second clip. Set against the closed models an ad team would otherwise rent, that is not a discount — it is a different order of magnitude. The same announcement prices Seedance 2.0 at $0.60 for a five-second clip delivered in a median 198 seconds, and puts Seedance 2.5's 720p image-to-video rate at about $0.473 per second in early September, which works out to roughly $2.31 for the same five seconds. That is a 12x gap against one comparable and a 47x gap against the other, and it is the reason this launch is worth more than a paragraph.

The second reason is rarer. Most launch posts sell a model and stop. This one carries a limitations section that concedes the resolution tier, names the clip type where post-training barely helped, and reports the benchmark that went worst. Those concessions are more useful to a buyer than the headline price, because they describe the jobs Boreal should not be given.

What Boreal is, in the vendor's own description

It is a video model for advertising specifically, generating both text-to-video and image-to-video at launch. Creatify states it is built on top of an open-source video generation model that produces audio and video jointly, then post-trained on a large body of advertising footage — the two failure modes the training targets are a product that stops looking like the product, and a presenter who stops reading as a real creator.

Two structural details are worth pulling out. First, Creatify says it owns the weights and serves them itself, which is what allows a per-second price rather than a per-credit one and removes the vendor queue that per-second competitors meter around. Second, output is 720p-class. The post is direct about that: where resolution-sensitive fidelity is measured, a 1080p flagship still wins, and Seedance renders inputs at 1080p against Boreal's 720p-class.

Speed is quoted as 1:1 with realtime — one second of finished video for every second of GPU time. The concrete version in the post is a twenty-shot storyboard finishing in under two minutes for about a dollar, against roughly twelve dollars and over an hour of queue on Seedance 2.0. Treat that as a vendor measurement of a vendor pipeline; it is consistent with the 198-second median the same post reports for Seedance, but it has not been reproduced outside Creatify.

The blind comparison, read past the headline number

The central evidence is a blind side-by-side of Boreal against its own untouched base model, holding prompt, resolution, frame count and seed fixed. Evaluators preferred Boreal in 81% of decisive comparisons — 25 to 6 with 9 ties across 40 production cases — and the post reports a sign test at p<0.001. Image metrics moved the same way, with a significant MUSIQ gain and fine-detail retention up 11.3% at p=0.004.

The breakdown is where the useful information lives. Broken out by case type the preference was 83% on product ads, 85% on creator and UGC scenes, and 60% on single-person talking clips. A model that doubles down on product fidelity and creator realism while improving least on talking heads is telling you what it was trained for, and talking-head video is the single most commoditized format in the ad stack — it is the one place where a cheaper model was already good enough.

Note what this comparison is not. It measures Boreal against its own base, not against the models an advertiser would actually choose between, and every number in it was produced by Creatify on Creatify's pipeline. The post is unusually explicit about that framing, and it also reports the separate head-to-head where a blind product-fair judge on 20 ad prompts put Boreal at 70% and Seedance 2.0 at 80% — a result that does not favour the model being launched. That kind of row is worth more than a clean sweep would have been.

Where the price claim needs a caveat

One cent per second is a list price at the 720p tier, billed per second of output rather than per credit or per token. Two things follow.

• The comparison is resolution-unequal — Boreal at 720p-class against Seedance at 1080p is a cheaper product being measured against a more expensive one, and the fidelity figures land within a few percent in both directions, which is the honest way to read it.

• Per-second billing removes queue time from your invoice but not from your schedule — the realtime claim is the part that matters operationally, and it is the part with the least independent evidence behind it.

Against H3 Max, the third reference point in the post, Creatify reports faster raw inference than Boreal's five-second target at roughly eight times the price per five-second clip at launch-day list. Against Seedance 2.5's image-to-video rate the gap reaches the 47x figure. Taken together these are three different claims of the same shape: the closed models are priced for advertisers who charge their own clients, and Boreal is priced for advertisers who iterate.

What you can call today

Boreal is live for Creatify users and reachable through the vendor's own API, its Model Playground, and its AdFlow agent, as well as through third-party platforms that host Creatify's models. It is not in our catalogue — OrcaRouter does not serve video generation models, and this article is not going to imply otherwise.

What the piece has to say about routing is narrower than usual, and still relevant. An ad pipeline is one video call wrapped in a lot of text work: brief expansion, script and hook variants, the compliance pass, the caption set, and the campaign logic that decides which cut runs where. That part of the stack is what we do serve, and it is where the cost of a campaign actually accumulates, because it scales with the number of variants rather than the number of clips. Those calls sit behind one key at provider list rate with 0% markup, and automatic failover keeps a variant-generation loop alive when a single provider degrades — which, on a pipeline that fires hundreds of small requests per campaign, is a scheduling concern rather than a theoretical one.

What to watch

Three things would change the picture, and none of them is another vendor comparison. The first is independent evaluation: Boreal's entire evidence base is a 40-case internal test and a judged 20-prompt bake-off, and no third party has run it. The second is whether the resolution tier moves — a 1080p Boreal at a cent-plus per second would take the main caveat off the table, and the roadmap in the post points instead at a defect-scoring judge and reference-conditioned generation for multi-shot work, which is the more interesting direction. The third is lip-sync, where Creatify says Boreal only matches its base and that a separate audio-specialized variant of the same lineage does better.

The defensible summary: the price is the least interesting thing about Boreal, because a per-second rate is easy to match and easy to explain. What is harder to copy is the combination of realtime rendering and a training target narrow enough to name its own weak spot. For short-form product and creator clips at volume, that is a genuinely new option. For anything that needs resolution, or a talking head, or an independent benchmark, it is not yet the obvious answer.