HeyGen + Google DeepMind in LA on June 11
A night of demos, conversations, and people building with agents, multimodal apps, and creative tools
Got something interesting? Lightning demo slots are open.
https://t.co/NWUmzjAtKg
yup - that's the bet we are making - thanks for recognizing us!
follow us @HyperFrames_ by @HeyGen
your agent writes code to build and edit videos, render into mp4, all local, all free.
open sourced here: https://t.co/g6XJPgZZvt
The most interesting visual AI tools today are generating the underlying source code behind the final output.
This change is unlocking editability, iteration, and a feedback loop that pixel-native models can't match.
And the market for visual code generation is organizing around the runtime where the artifact is rendered or executed.
a16z's Yoko Li on why the next frontier of visual AI is code: https://t.co/tIA8luD4OG
thanks for featuring us @a16z!
code-to-visual is a trillion $$ market
every clip in this montage is code (and the video itself)
it's still super early for @HyperFrames_ , so much to build!
hey builders! showcasing your code & CLI w @HyperFrames_ got 10x easier
We open sourced accurate, pre-built UI components
A Mac terminal. A VS Code window. Free.
One command to install
$ npx hyperframes add code
Swap in your own code or commands and go
This is how I source B-rolls using Claude code for my AI-generated videos
And here's when each one makes sense.
1. HTML @HyperFrames_ – for "viewer reads on-screen text" cuts (free)
Claude renders these in a browser. Very deterministic.
2. @pexels – for stock filler b-roll (free)
Clocks, kettles, hands pouring and other stock elements. Returns the mp4 directly.
3. Nano Banana still – for product-in-context shots
I take stock photos and have Claude Code composite products on with NB or GPT Image 2.
4. Nano Banana/GPT Image 2 + Sora 2 – for motion on a stable subject
I generate images using NB2/GPT Image 2 and hand it over to Sora 2. Use this for product hero shots and lifestyle context with motion. This combination yields amazing product consistency.
5. Seedance 2.0 ref-to-video – for talking-head UGC
Pass a reference image and lip-synced VO, get up to a 15s multi-scene clip with a person speaking.
6. Veo 3.1 i2v – for physics-driven motion
Water, fabric, smoke, splash. If the motion doesn't need physics, source 4 does the job.
For each cut, Claude Code picks based on what the moment needs to communicate.
These learnings are distilled from our experience creating hundreds of ad videos for clients.
Hope you find them useful.
HyperFrames is an open-source framework that turns HTML, CSS, and animations into deterministic MP4 videos, built for AI coding agents and automated pipelines.
- Define videos as HTML with data attributes for timing and tracks
- Works with GSAP, CSS, Lottie, Three.js, and other animation libraries
- CLI for local preview with live reload and MP4 rendering
- Agent skills for planning, writing, linting, previewing, and rendering videos
Many financial advisors spend hundreds, sometimes thousands, per month on client update videos
We wanted to see if Avatar 5 + HyperFrames could handle the entire task
So we gave an AI agent a brief and let it run
Here's what it created, plus the workflow in the thread below ↓
Parallax unzoom effect on @HyperFrames_ thanks to the community building this, available to use now👇
Reply with more cool effects you want in the catalog that don't exist yet but they would be a banger, we'll try to make them, or even you can try to make one
I use this to do my biweekly engineering updates for @HeyGen. It pulls my github activity, uses my avatar 5 with the heygen CLI (it can be created too), and renders the video with @HyperFrames_ CLI.
The whole thing runs from Claude Code/codex/hermes agent with one skill in the reply below 👇
the @claudeai team really cooked 👏
this hyperframes video was made by Opus 4.8 (directed by @Miguel07Code)
really clean camera movements and visual aesthetics