September 30, 2026
Video ads follow briefs more closely: Boreal-H3 fulfilled half of the complex briefs
Creatify fine-tuned MiniMax H3 for advertising: the model is designed to keep products and characters consistent and check on-screen actions against the brief.
🚨 AI News | TestingCatalog
@testingcatalog
Creatify launched Boreal-H3, a video model for ad production, and Ad Agent powered by Claude Opus 5.5. Boreal-H3 was fine-tuned from MiniMax H3 using real advertising projects. On the same tasks, Creatify measured an increase in the share of completed briefs from 28% to 50%, identity preservation from 83% to 94%, and a 71% reduction in error severity. > Products retain their shape, labels, and details as the camera moves around them. > Generated characters handle user-style ads, complex spoken lines, and dialogue between two people. > Users can specify camera movements and keyframes, including for product interactions and physical actions. Ad Agent created the ad below in a single run.
Introducing Boreal-H3 — a video model built for ads and our next step toward recursive self-improvement in video generation. A good-looking video isn’t enough. The product has to stay the same. The actor has to stay the same. The label has to be right. And the action in the brief actually has to happen. So we post-trained MiniMax H3 specifically for advertising. But this isn’t a one-off SFT or LoRA fine-tune. We built a closed-loop system that learns what to improve next. Human-calibrated evaluation diagnoses failures and guides the next intervention: targeted data collection, reinforcement learning, or inference optimization. When the feedback is unreliable, we revise the evaluator or reward—not just the generator. Every experiment feeds into shared memory, informing the next training decision. The model improves, and so does the process that produces its successor. The results: → 85.3% reference fidelity — highest among the frontier video generation models we evaluated → Brief success: 28% → 50% → Identity match: 83% → 94% → Visible defects per clip: down 70% → Generation time and estimated cost: down 20% Boreal-H3 doesn’t just make better-looking video. It makes more usable ads. Credit to the @MiniMax_AI team for the foundation we’re building on. This launch is a checkpoint, not the finish line. We’re building more than a better video model. We’re building a system that learns how to make the next one better.
· 5.8K views
Boreal-H3 completed 50% of complex advertising tasks in Creatify's internal evaluation from Sep 28. The base MiniMax H3 completed 27.8%.
Creatify trained the model on real advertising projects. Ads need to preserve product shapes and packaging text, keep characters recognizable, and carry out the actions specified in the brief.
What changed. In the same evaluation, character consistency rose from 83.3% to 94.4%. Creatify compared the base H3 with the full Boreal-H3 pipeline, including prompt processing. The number of visible defects per clip fell from 1.11 to 0.33, or by 70%. Separately, the company reports a 71% reduction in error severity.
The model supports customer-style ads, complex spoken lines, and dialogue between two characters. Users specify camera movements and keyframes, including for product interactions and physical actions.
How to use Boreal-H3
Access in Creatify. In Model Playground, select Image to video and upload a starting frame. You can add an ending frame if needed. Boreal-H3 Reference mode combines product, person, and scene references, as well as video and audio.
| Parameter | Terms as of Sep 29 | | --- | --- | | References | Up to 9 images, 3 videos, and 3 audio clips | | Generation at 768P | $0.04 per second of video | | Generation at 1088P | $0.12 per second of video | | Generation at 2K | $0.26 per second of video | | Modes at this price | Text to video, Image to video, Reference to video |
According to Creatify's measurements from Sep 28, a 10-second video at 768P costs $0.40 and takes about 11 seconds to generate. In a comparison example from Sep 29, the company measured the time to generate a 15-second video featuring a squishy toy.
| Model | Generation time in Creatify's example | | --- | --- | | Boreal-H3 | 84 seconds | | Seedance 2.5 | 411 seconds |
Alongside the model, Creatify is turning Creative Agent into Creatify Ad Agent, powered by Claude Opus 5.5. The AI agent develops ideas, produces ad variants, and checks the results. The company produced its demo video in a single agent run.
Creatify is building a model improvement loop: error evaluation determines which data and training are needed next.
