CEO @Creatify_AI. Building AI that creates & optimizes ads. Previously Snap + Meta. Sharing what we learn building AI models and generating ads at scale.
Mountain View, California
Joined March 2012
- Tweets5.8K
- Following518
- Followers3.2K
- Likes17.5K
Pausing losers and scaling winners isn't agentic. The platforms already do that.
Agentic starts when the agent writes the next batch of ads from what just won. Budget is the output. Creative is the input. nitter.cf/rokhladnik/status/2111…
I’m seeing more and more brands falling for the "AI media buying" pitch.
If your AI tool is:
- Increasing or decreasing Meta/Google/TikTok budgets based on yesterday's, L3D or L7D performance
- Pausing underperforming ads
- Scaling campaigns when ROAS hits a certain threshold
Congratulations. 🙌
You’ve basically discovered automated rules. 🙃
Something advertisers have been using for years.
And there's nothing wrong with automation. We use it too.
But let's not confuse automation with intelligence.
What bothers me is how these tools are being sold.
Someone mentions "AI" or "superintelligence" and suddenly brand owners start calculating how many media buyers they can fire.
Meanwhile, the tool is just moving budgets up and down based on historical performance.
Or even worse: just making suggestions what to scale or downscale and someone has to manually do it.
That's not necessarily better optimization. And it certainly doesn't guarantee better business outcomes. Which is what you should focus on.
Before paying a premium for an "AI media buyer" ask:
What is it actually doing that wasn't possible before?
What decisions does it make better than an experienced operator?
And most importantly, can it demonstrate incremental business results?
Don't pay AI prices for automated rules.
Thanks.
Have a great weekend.
Steven Yinan Na retweeted
Same script. Same talking-head ad. Aurora 2.0 vs VEED Fabric 1.0.
A 15s clip costs $1.05 on Aurora 2.0 ($0.07/s) vs $2.25 on Fabric 1.0 ($0.15/s, 720p).
2.1x cheaper, and a higher Avatar Performance Score: 90.5 vs 87.4.
Learn more: labs.creatify.ai/models/auro…
Steven Yinan Na retweeted
Boreal-H3: the video model built for product ads. 1088P at $0.12/sec.
Drop in up to 9 product photos. Get one 360° video, front to back. No turntable. No reshoot.
Video model will become specialized just like LLM does today.
Better and cheaper than its own base.
Boreal-H3 is the only model in the top-left: 133.3 on ads quality vs. MiniMax H3's 100, at half the price per second.
Read more: labs.creatify.ai/research/bo…
Steven Yinan Na retweeted
Built for ads, scored on ads.
Boreal-H3 tops our Ads Quality Index at 133.3 (MiniMax H3 = 100).
Creatify internal evaluation, Sep 2026, same briefs for every model.
Full table and method → labs.creatify.ai/research/bo…
Excited to launch a home base for @Creatify_AI Research, where we'll collect and share all of the work we're doing around model training, ads quality benchmark and research collaboration.
Introducing Creatify Labs.
Building better ad agents requires training specialized models, rigorous evaluation, benchmarks grounded in real ad production and campaign results, and research collaborations that push the frontier of video.
We’re sharing our research agenda and our work across each of these areas, starting with Boreal-H3, our video model trained for ads, and Aurora 2.0, our avatar model built for UGC ads.
Live at: labs.creatify.ai
The future is open-weight. congrats @fishaudio team
$52M seed round. 15M+ users. But this is what we're most proud of yet.
Introducing Drama 3: the most emotive and controllable TTS model in the world
And yes, it's open-weight.
You can now control how even a single word should sound, change its tone, and shift your voice mid-sentence.
Uncle Roger was impressed too.
Try it here:
s.fish.audio/rqh0l8
Comment “DRAMA” within the next 48 hours for 1 month free.
If you're an enterprise your first month is on us:
s.fish.audio/qz9r0t
Steven Yinan Na retweeted
14 shots. Every one generated with Boreal-H3.
Total cost: $2.80. Ad-quality video at $0.04 a second.
That's ~12× less than Seedance 2.5.
Steven Yinan Na retweeted
It is in early access, however Creatify’s Ad Agent is live at creatify.ai.
Their headline says 10× the creative volume and 2× ROAS, but they haven’t shown that one caused the other.
How many creatives actually received meaningful spend? What changed in targeting, retargeting and attribution? Until that’s disclosed, it’s a before-and-after marketing story.
In the demo, it's thru moving prospecting to retargeting, ofc that "works" ;)
Introducing Runway Ads, an autonomous engine for performance marketing.
Since July, we’ve used Runway Ads to:
-Scale ad volume 1,000% while doubling our return on ad spend
-Increase conversion by 34% with click-through rate holding steady
-Bring our cost per subscriber down 41%
-And add $100M ARR
Now, we’re making it available to select enterprise partners and will be rolling it out widely in the coming weeks.
Learn more and request early access: runway.com/news/company-news…
Steven Yinan Na retweeted
4/
The visual examples show the difference. The evaluation tells the same story.
→ 85.3% reference fidelity — highest among the frontier models evaluated → 133.3 Ads Quality index — highest observed across six frontier models
Ads Quality jointly measures reference consistency, prompt following and visual realism.
We optimize for whether a generation is actually usable as an ad—not just whether it looks good.
What happened to language model will happen in video. The era of open video model is here.
Built on MiniMax H3, @Creatify_Labs' Boreal-H3 is a video model optimized for advertising, keeping products and characters consistent while following creative briefs more accurately.
Excited to see MiniMax H3 serve as the foundation for more frontier models tailored to specific industries! ✨
Steven Yinan Na retweeted
Really excited to finally share Boreal-H3, our video generation model built for advertising.
What excites me most isn’t just the model quality—it’s our post-training approach behind it.
We started with a strong foundation model, then built a feedback-driven post-training system around the things that actually matter for advertising: reference fidelity, brief following, identity consistency, motion, and failure detection.
Evaluation feeds back into what data we collect, which failures we target, and how we design the next round of post-training—so each cycle improves not only the model, but the process for improving it.
The result: Boreal-H3 reaches 85.3% reference fidelity while making major gains in ad usability.
This is a strong example of how much capability can still be unlocked through post-training—and a step toward a more systematic, self-improving paradigm for video generation.
Huge credit to the team behind this. More technical details in the launch post 👇
Read our technical report: labs.creatify.ai/research/bo…
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.
Today we launched Boreal-H3, our new video model purpose-built for advertising.
We’ve spent the last few years building AI for ads, and it’s led me to a broader belief:
Ads are to video intelligence what code is to language intelligence.
Coding became one of the best environments for training language models because it has everything AI needs to learn: high-value work, fast iteration, objective feedback, and constant corrections.
Advertising has the same properties for video.
You create an ad. Judge it. Run it. Measure performance. Revise it. Repeat. Every cycle produces creative decisions and real-world outcomes like CTR, CPA, and ROAS.
That’s the bigger idea behind Boreal-H3. We’re not just trying to build a model that generates better-looking videos. We want to build video intelligence that learns from the full loop of creation → judgment → performance.
Code helped turn LLMs into reasoning engines.
We believe ads can do the same for video intelligence.
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.
Faster, cheaper, better - our best avatar model yet
Introducing Aurora 2.0 —our best avatar model 🔥
One photo. Your exact voice. Up to 60s of expressive video. Sharper lip sync. Natural gestures. 2× faster than Aurora. $0.07/s.
🏆 Highest overall score in our avatar benchmark.
Build UGC ads, podcasts, product explainers, and character videos. Swap the voice track to test a new hook, change the offer, or take your campaign into another language.
Steven Yinan Na retweeted
Now you can access Creatify Aurora model through @ElevenLabs API 🤝
