AI Engineer (Agentic Ad Creative (Multimodal))

Join NewsBreak as an AI Engineer, where you'll be at the forefront of revolutionizing the creative pipeline for digital advertising. You'll own the end-to-end development of an agent that generates thousands of images and videos daily across multiple platforms. Your work will directly impact the efficiency and effectiveness of our advertising efforts, making it a crucial role in our growth strategy.

• Extend the CPI-feedback creative critic loop to include images, video, and copy, running unsupervised at thousands of variants per day.
• Build a video pipeline that generates platform-ready ads from a topic and product brief in under 90 seconds per iteration.
• Develop a generative model router that selects the appropriate model for each brief, minimizing cost per asset while maintaining quality.
• You understand diffusion and video models viscerally — you’ve debugged them, trained a LoRA, optimized inference, built a ComfyUI workflow you’d defend in public
• You have public artifacts — a fal demo, a Replicate model, a Civitai upload with >100 downloads, a viral X thread of your generations, a ComfyUI workflow on GitHub. Portfolio matters more than resume
• You have scroll-stopping instinct — you know what makes someone not swipe past an ad in the first 3 seconds
• You’ve read Meta Advertising Standards / TikTok branded-content rules end-to-end — policy is a pre-deploy gate, not a post-mortem
• You think about cost as a first-class design constraint — $/asset, $/click, $/install. Cost arbitrage between model tiers is something you’ve actually shipped
• You can name your daily model stack — Flux vs SDXL vs Veo vs Sora vs Runway vs Kling vs Hailuo — and explain when each wins, where each breaks, and what you’d reach for at $0.01/asset vs $0.50/asset
• You’re a builder — Idea → working pipeline → first 100 ads shipped → CPI data back → tighter loop. Not “I’ll spec it for Q3.”
• Pretrained or significantly advanced a VLM (not just SFT’d or LoRA’d one)
• Open-sourced a ComfyUI custom node, a diffusers extension, or a Replicate model with real traction
• Deployed serverless inference on fal / Replicate / Modal at scale (>10k req/day)
• Shipped a video product end-to-end: script → scene plan → b-roll → voiceover → captions → final cut
• Authored a public take on generative video model failure modes that other engineers cite