Creatify Labs today launched Boreal-H3, an AI video model for ads built by post-training MiniMax H3 in partnership with MiniMax. Creatify trained the model on the work an ad has to do: keep the product from the reference image intact, deliver natural creator performances, and carry out the brief as written.
Boreal-H3 scores a third higher than its MiniMax H3 base on Creatify's Ads Quality Index and costs $0.04 per second, about one-twelfth the price of Seedance 2.5.
The Ads Quality Index measures how often a model produces a clip that works as an ad. A clip passes only when prompt following, reference consistency and visual realism each score 6 out of 10 or higher, and each model's pass rate is indexed against MiniMax H3's. Boreal-H3 scored 133.3, ahead of Omni 1.1 Flash at 122.2, Seedance 2.5 at 118.1 and Wan 3.0 Prime at 111.1.
On reference fidelity, scored at a stricter 7 out of 10 threshold, Boreal-H3 reached 85.3%, ahead of MiniMax H3 at 82.4%, Wan 3.0 Prime at 79.4%, Seedance 2.5 at 71.9% and Omni 1.1 Flash at 70.6%. Creatify's automated judges were validated against human preferences, and its human studies used anonymized, randomized outputs.
The product stays the product, and the creator stays the creator. Against its MiniMax H3 base, Boreal-H3 raised subject consistency from 83.3% to 94.4%, keeping the product and the creator recognizably the same across a clip. On a set of difficult production briefs, it raised brief completion from 27.8% to 50.0% and cut visible defects from 1.11 to 0.33 per clip, a 70% reduction.
The model is also built for the performance side of a UGC ad. Boreal-H3 handles complex dialogue and multi-avatar conversations, with a focus on clear speech, natural pacing and a creator who can handle the product while speaking. Product demonstrations get the same attention, with cause-and-effect actions such as a vacuum leaving a clean path or a pour that behaves like liquid.
Boreal-H3 gives creative teams two ways to direct a shot. Keyframe control, in image-to-video, preserves the subject and visual style of an opening frame, with an optional ending frame to set exactly where the shot lands. Multi-reference control, in reference-to-video, combines up to 9 images, 3 videos and 3 audio clips into a single coherent shot. Text-to-video is also supported, and the model follows camera direction with natural transitions between scenes.
Boreal-H3 generates video at 1.1 seconds per video-second at 768p, so a 10-second clip takes about 11 seconds and costs $0.40. In the same comparison, Seedance 2.5 ran at 46.5 seconds per video-second and about $0.47 per second, which makes Boreal-H3 roughly 42 times faster and 12 times cheaper.
Creatify notes that the evaluation measures ad quality before launch, and real campaign testing still has to determine which ads convert.
Pricing and availability
Boreal-H3 is available now in Creatify's Model Playground and through the Creatify API, with text-to-video, image-to-video and reference-to-video modes.
About Creatify Labs
Creatify Labs is the research lab building lifelike AI video models, including the Boreal family and the avatar model Aurora.
