Before / raw clip
23.47 seconds · abandoned opening kept
A local-first editing workspace for Codex and Claude Code
Turn one raw talking-head clip into a finished Meta ad.
Give your coding agent the video. It finds the clean take, conforms the captions, adds restrained hook and CTA graphics, checks the render, and opens a reviewable MP4 before anything is final.
Apple Silicon Mac · Codex or Claude Code required · local preview before approval
Awaiting your review1080×1350 · H.264/AAC
Local media processingNever overwrites originalsReview before approvalResolve handoff optional
Rendered from the shipped workflow
See the raw clip and finished ad side by side.
The demo uses a clearly disclosed synthetic presenter and system voice so the complete proof package can ship without exposing customer media.
RAW
After / review preview
17.3 seconds · cut, captions, hook, CTA, QA
Transparent demo disclosure: the fictional presenter image and system voice are synthetic. This is product proof, not a customer testimonial or an ad-performance claim.
8/8
Required doctor checks
Passed from a fresh extracted buyer ZIP on the tested Apple Silicon Mac.
24
Automated checks
Unit and integration coverage plus a real opt-in HyperFrames render gate.
−14.39
Demo dialogue LUFS
Measured from the distributable finished proof, not an ad-performance metric.
0
HyperFrames warnings
Strict runtime, layout, motion, and contrast checks passed on the finished proof.
One plain-language prompt
Tell the agent what “done” means.
You do not configure an MCP server or hand-edit a timeline to get the first result. The included skill owns the technical sequence and asks for the commercial facts it cannot know.
Your first prompt
“Turn this video into a 4:5 Facebook ad. Preserve my meaning, do not invent claims, and open a preview before finalizing it.”
01 / LISTEN
Transcribe the recording
02 / CUT
Build an inspectable edit plan
03 / DESIGN
Add captions, hook, and CTA
04 / REVIEW
Check the output and wait
Three steps to the first preview
No timeline apprenticeship required.
- 01
Choose the clip
Drag in a video you own. The original is hashed and never overwritten.
- 02
Answer three business questions
Tell the agent the offer, audience, and desired action. Add permitted evidence and claims to avoid.
- 03
Review before approval
Inspect the preview, transcript, cuts, captions, graphics, QA, and uncertainty. Revise or approve it yourself.
Inside the core
The working editing system—not a prompt PDF.
For founders, marketers, consultants, and small teams who already have Codex or Claude Code and want their own local, inspectable talking-head editing workflow.
CORE / 01
Agent-owned Mac setup
Pinned dependencies, local environments, browser preparation, a product-specific doctor, and a deterministic demo.
CORE / 02
Truthful editorial planning
Local transcription, source-backed hooks and beats, validated cut ranges, confidence markers, and prohibited-claim guardrails.
CORE / 03
Guarded clean-cut rendering
Kinocut where reliable, expected-duration verification, and an automatic timestamp-reset FFmpeg fallback when the render drifts.
CORE / 04
Reusable HyperFrames finishing
Social framing, output-timed captions, restrained hook and CTA graphics, motion assertions, contrast checks, and sampled-frame review.
CORE / 05
Dialogue and delivery QA
Approximately −14 LUFS dialogue normalization, H.264/AAC yuv420p checks, duration checks, and human-review limits stated plainly.
CORE / 06
Review, resume, and handoff
A local review page, hashed artifacts, one brand profile, SRT, receipts, support bundle, flattened master, and optional simple FCPXML.
Review receipt / proof-4x5
Human review: pending
- Expected / actual
- 17.3s / 17.3s
- Dialogue loudness
- −14.39 LUFS
- Delivery
- H.264 / AAC
- HyperFrames gate
- 0 warnings
Automated QA does not pretend to prove semantic accuracy, caption taste, safe crops, or ad performance. Those stay visible for human review.
Inspect, then decide
A finished file is not the same as an approved ad.
The local review page shows what was cut, low-confidence decisions, captions, graphics, evidence, duration, codecs, loudness, and output files. The agent can prepare the preview. Only you can approve it.
Three production modes
Start fast. Add complexity only when it earns its keep.
Default first preview
Fast
Clean cuts, social framing, accurate captions, dialogue normalization, and restrained hook and CTA treatment.
When evidence needs explaining
Designed
Adds supplied screenshots, comparison cards, useful callouts, and proof annotations that trace to real evidence.
Opt-in, slower compositing
Studio
Adds cached subject matting or a virtual environment only when the footage and extra render time justify it.
Local media by default
Your footage is not an excuse for vague privacy copy.
Transcription and rendering run locally. The product never uploads or publishes media by itself. Codex or Claude may still inspect project context under that provider’s terms, so the documentation says that plainly.
The original stays immutable
Every job records the source path and hash, then writes new artifacts into an isolated job directory.
Support bundles stay narrow
The generated report excludes footage, transcripts, credentials, and unnecessary personal paths by default.
No invented claims
Source meaning, qualifiers, prohibited claims, and unsupported quantitative overlay claims remain guarded.
No silent publishing
Rendering a preview is allowed. Approval and ad-platform publishing remain human decisions.
Compatibility and fit
Buy it for the machine and workflow it actually supports.
Good fit
- ✓ You have an Apple Silicon Mac.
- ✓ You already use Codex or Claude Code.
- ✓ You own or can legally edit the footage.
- ✓ You want inspectable local files and a human review gate.
- ✓ You are comfortable letting the agent manage local dependencies.
Not the launch version for you
- × You need Windows, Intel Mac, Linux, or a phone app.
- × You need an automatic ad publisher.
- × You expect the tool to invent proof or performance claims.
- × You need custom editing or creative strategy included.
- × You want a notarized desktop GUI instead of an agent workspace.
Core product
Make the first reviewable ad for $27.
The core is useful by itself. Setup, review, resume, caching, QA, captions, receipts, and one brand profile are not held back for an upsell.
Compatibility guarantee
If the included demo will not render on a Mac we list as supported, you should not be stuck with it.
Send the generated doctor report within 14 days. If the included setup and demo cannot be made to work on the listed configuration, request a refund. Support covers the included setup, doctor, and demo—not custom editing, operating-system repair, ad strategy, or publishing.
One-time price
$27USD · one time
- ✓Apple Silicon Mac setup and product doctor
- ✓Codex and Claude Code launchers and agent skill
- ✓Kinocut, HyperFrames, FFmpeg, and whisper.cpp workflow
- ✓Fast, Designed, and Studio production guidance
- ✓Review page, resume, one brand profile, QA, and receipts
- ✓Flattened MP4, SRT, edit plan, and optional simple FCPXML
The deliverables are built. Checkout stays disabled until Dodo test products and entitlement downloads pass end to end.
Optional after checkout
$9 Hook & Recording Pack order bump · $37 Ad Test Lab upsell · neither is required for the first finished ad.
Before you buy
Clear requirements. No editing mystery box.
Does this replace DaVinci Resolve?
No. The core path is agent-operated Kinocut, HyperFrames, and FFmpeg. The reliable handoff is a flattened master, SRT, edit plan, and receipts. Simple FCPXML is a best-effort bonus.
What computer does version 0.1 support?
Apple Silicon Mac is the launch target. Codex or Claude Code is also required. Windows, Intel Mac, and Linux are not advertised until they pass clean-environment acceptance.
Is all processing offline?
Transcription and rendering are local by default. Your coding agent may inspect prompts, filenames, transcripts, or project files under that provider’s terms. Optional cloud services require separate approval.
Will it publish my ads?
No. It may prepare and render previews, but it never publishes an ad and never approves the preview for you.
Can it invent a stronger claim?
It must not. The product preserves source meaning, checks prohibited and unsupported quantitative overlay claims, and shows uncertainty for review.
What is the Hook & Recording Pack?
A $9 optional order bump for recording the next source clip more effectively. It is not required to make the first ad.
What is Ad Test Lab?
A $37 optional post-purchase module for organized hook, CTA, style, and format variants after one standalone ad already works.
Is the person in the demo a customer?
No. The presenter image and voice are synthetic demonstration media. The workflow artifacts and measured technical checks are real; no testimonial or ad-performance claim is implied.
Your next talking-head ad can start with the clip you already recorded.
Give your agent the footage, the commercial facts, and the instruction not to invent claims. Review what comes back before anything is final.
Get the Ad Machine — $27Apple Silicon Mac · Codex or Claude Code required · local preview before approval