From the first idea to the finished ad.
Choose the product, audience and message. Hadron turns the brief into video, image or slideshow ads. Compare every version side by side and choose the winner.
One process for every step behind the ad.
Hadron brings creation, review and export into one production process, so teams no longer have to coordinate each stage across separate tools.
Most AI tools deliver only the finished file. Hadron also keeps the production history behind every version, including the brief, source files, reviews, costs and final export. This makes it clear how each result was produced.
For teams that produce and compare ad variations.
A limited number of companies are currently using Hadron in private beta. New companies will be added by invitation.
DTC and e commerce teams, as well as performance marketing studios that want to produce more ads without managing every prompt, tool and file themselves.
Nineteen points form the Hadron mark.
Seven points form the H and twelve create the surrounding field. The geometry gives the mark a precise center and a lighter outer form.
| Grid | 64 unit box with an 8 unit major module |
| Cloud | Twelve points on a radius of 28, every 30° |
| H columns | x=22 and x=42 |
| H rows | y=20, y=32, y=44 |
| Lockup | Mark 32, gap 10, wordmark 18, weight 600, tracking -0.01em |
Keep one 8 unit module of clear space around the mark. No text, icons or edges may enter this area.
Whitespace separates information and controls the pace.
Structure comes before styling. Every element has a defined role.
Exact labels and values. Anything that doesn't help the user decide is removed.
The visual identity is structured and technical, with emphasis on information, status and comparison.
Tokens define color, type, spacing and motion.
Each visual role has a named token. The UI uses these tokens through CSS custom properties and Tailwind v4's @theme block in globals.css.
| tile | 12px |
| card | 14px |
| pill | 999px |
| shadow | Three steps: xs, sm, md |
| ease-smooth | 0.22, 1, 0.36, 1 |
| ease-swift | 0.4, 0, 0.2, 1 |
| dur-fast | 120ms |
| dur-base | 180ms |
Each easing and duration is a token. Replay the demo to compare them.
Use fast for feedback, base for movement and slow for surface changes.
Use rise for entry, fade for replacement and a 60ms stagger for sequences.
Controls in the app use 36. Landing, auth and public pages use 48. The two densities stay separate by design.
From draft to export, every state is visible.
Key product states are rendered from the product code, with supporting screens defined below.
The V1.1a pipeline runs from brief to final export. Each run moves through seven steps, with the status, cost, review and files saved along the way.
| Step | Runs | Per run | Total |
|---|---|---|---|
| Script | 4 | $0.04 | $0.16 |
| Images | 4 | $0.12 | $0.48 |
| Video | 4 | $0.14 | $0.56 |
| Captions | 4 | $0.01 | $0.04 |
| Total | 4 | $0.31 | $1.24 |
The website explains what Hadron does, how it works and how to request access.
The system connects the web app, queue, storage and AI providers.
Current system architecture.
| Deployment | Vercel for the frontend and Google Cloud Run for the worker |
| UI Components | Radix Dialog, Sonner and custom token based components |
| Frontend | Next.js 15 App Router, React 19, TypeScript in strict mode |
| AI Providers | Anthropic, OpenAI, fal.ai, Google Vertex AI and Replicate |
| Platforms | Responsive web app for desktop and mobile devices |
| Analytics | Vercel Analytics, Speed Insights and PostHog EU |
| Authentication | Supabase Auth with a Google OAuth allowlist |
| Database | Supabase Postgres with Row Level Security |
| Styling | Tailwind CSS v4 with @theme tokens |
| Queue | pg-boss on Supabase Postgres |
| Worker | Node.js on Google Cloud Run |
| Fonts | Geist and Geist Mono |
| Storage | Supabase Storage |