How to Make Launch/SAAS Videos with HyperFrames + GPT-6 Astra

How to make stunning videos with HyperFrames driven by GPT-6 Astra point it at any video you love, hand it your product, and it rebuilds that same feel and renders a real MP4. One prompt to paste, one optional polish pass, and the workflow explained plainly enough to teach someone else.

</aside>

The whole idea in one line

Find a video whose look and rhythm you want. Hand that video plus your product to GPT 6 Astra. It measures the reference frame by frame, rebuilds the same structure as HyperFrames HTML, generates the visuals you don't have, voices it with ElevenLabs, mixes it, and renders the MP4.

<aside> 👉🏻

Bring your own reference. Any video works a SaaS promo you like, an Apple-style product film, a competitor's launch video, or something you shot yourself. The examples on this page used this video saasvideo.mp4, which is my reference video.

</aside>

What is HyperFrames?

HyperFrames turns a webpage into a video. Your agent writes the scenes like a webpage, and the computer records it frame by frame into a finished MP4 — no editor, no timeline, no dragging clips around.

What you need: a coding agent running GPT-6 Astra (Codex, Claude Code, Cursor any of them), the video you want to copy, and an ElevenLabs key [ optional ] for the voice and SFX sound.

Fill these in when you paste

Placeholder	What goes there	Where it comes from
`[ reference video path ]`	Full path to the video you want to match. Atharv's example: `C:\Users\Desktop\athrix-Codex\Project1\saasvideo.mp4`	You
`[ your product name ]`	Whatever the video is for a SaaS, an app, a model launch, a single feature	You
`[ your ElevenLabs API key ]`	The key , you wanna give to ai agent	Elevenlabs.io API Key
`[ one line on what it is ]`	The positioning line for your product, 12 words or more	You
`[ feature 1 ]` `[ feature 2 ]` `[ feature 3 ]`	The three things this video absolutely has to land	You [ optional ]
`[ voice name or ID ]`	The narrator [ optional ] or let the agent shortlist three and pick	ElevenLabs Voice Library
`[ what extra that you need ]`	Leave this in. It forces the agent to list its gaps instead of quietly guessing	The agent fills it

Prompt for video

We are making a launch video for [ your product name ] with HyperFrames — <https://github.com/heygen-com/hyperframes>
You are GPT-6 Astra. Run the whole loop yourself: watch the reference, build it, review your own frames, fix it, hand me the MP4.

SETUP
npx skills add heygen-com/hyperframes (core skills only). Read /hyperframes first, then enter /product-launch-video.

REFERENCE — match this, do not reinvent it
[ your reference video path ]
Probe it, pull frames every 0.5s, and actually look at them. Write the shot list into frame.md: every cut timestamp, pacing, camera moves, easing, text placement, palette in hex, type weights, where the beats land. Then rebuild that same structure and rhythm for my product — same energy, new content. It should feel like the same editor cut both videos.

PRODUCT: [ your product name ]
- What it is: [ one line on what it is ]
- Hero features: [ feature 1 ], [ feature 2 ], [ feature 3 ]
- Tone: [ calm / premium / cinematic / hype ]
- Brand: logo [ path, or "generate one" ], colours [ hex ], type [ font, or "you pick" ]
- End card: [ "your product — available today" + site or handle ]
- Specs: match the reference for size, fps and length. Output to exports/[ project-name ].mp4

ASSETS
Generate everything I have not given you with Image Gen — mockups, product shots, backdrops, icons, logo lockups, end card. One consistent look across all of them, 2x for anything that scales or moves. No placeholders, no lorem ipsum, no empty frames.

AUDIO
- ElevenLabs key: [ your ElevenLabs API key ] — use it for the voice and the SFX calls
- Write the narration script and show me before you spend credits: [ 90 ] words max, one idea per scene, ends on the CTA
- Voice: [ voice name or ID, or shortlist three and I will pick ]. Generate per scene, then transcribe for caption timings.
- SFX from <https://elevenlabs.io/sound-effects> — transition whooshes, reveal ticks, one riser or braam on the logo hit, a clean tail on the end card. Prompt them in audio terms, keep them tight, land them exactly on the cut.
- Music: [ music file path, or source one that matches the reference energy curve ]
- Mix with /hyperframes-audio: voice on top, music carved under it, limiter on the master, around -14 LUFS

BEFORE YOU CALL IT DONE
Lint, check, snapshot, and review your own frames. Then render and report the output path, the duration, and anything you had to compromise on.

Brief me back in 5 lines before you start. If anything is missing, ask it all in one batch — do not guess on brand or copy.

WHAT ELSE DO YOU NEED FROM ME?
[ what extra that you need — list it in one go ]

Optional prompt - the polish pass

<aside> ✨

Skip this if the first render already looks right. Use it when the video is close but not exact it makes Astra critique its own work before you have to.

</aside>

Compare your render against the reference before I watch it.

1. Snapshot both at the same timestamps and put them side by side
2. Tell me honestly where ours is weaker — pacing, contrast, spacing, type, transition timing, audio balance
3. Rank the top 5 fixes by impact, fix the top 3, re-render

Only touch what you listed. Do not redesign scenes I already approved.

How the workflow runs

Five steps. You do the first one, Astra does the rest.

You paste the prompt. Astra installs the HyperFrames skills and picks the right workflow.
It reads your reference. Probes the file, pulls frames, writes the shot list into frame.md.
It builds. An index.html of timed clips and GSAP timelines, plus every asset, voiceover and SFX it needs.
It checks itself. Lint, snapshot, look at the frames, fix what is off.