Blog
Notes from the canvas
Product news, engineering deep dives, and the prompting playbooks we use ourselves.
Prompting video models in 2026: what actually works
Veo, Kling, Seedance and Hailuo all read prompts differently. After a few hundred thousand generations, here’s the playbook we give our own team.
Read the post
How we made AI generation survive anything, with Backblaze B2
A video job can take minutes and cost real money. Here’s the durable job pipeline that means a crashed server, a closed laptop, or a redeploy never eats your generation.
MediaMint Engineering · 9 min read
Credits, explained: one honest price for every model
One credit is $0.002 of raw inference — for every model, every plan. Here’s the math, why video is priced per second, and how to stretch a balance further.
The MediaMint Team · 6 min read
Making an infinite canvas multiplayer (without losing anyone’s work)
Live cursors are a demo; not destroying each other’s edits is a product. How MediaMint syncs boards in real time — presence, merge rules, and an offline cache.
MediaMint Engineering · 7 min read
Why every workspace gets its own budget
Shared team balances always end the same way: one project eats the month by the 19th. Per-workspace credit pools are our answer — here’s how studios run them.
The MediaMint Team · 5 min read
Introducing MediaMint: every model, one canvas
The best image model, the best video model, and the best audio model are made by different companies — and your project shouldn’t care. MediaMint is live today.
The MediaMint Team · 5 min read