Codex edits a fishing montage through phone review
Makelky reports a published 31.7-second fishing montage assembled by Codex from real footage and ten photos, revised through mobile MP4 reviews and delivered in two export versions.
aaronmakelky.com
Original prompt
Shared by the source author. See source notes for context.
I'm on mobile gpt remote, I can't see a localhost video. Share it with me in a medium I can view here
Reach people exploring AI agents
See placements →What happened
Makelky reports a published 31.7-second fishing montage assembled by Codex from real footage and ten photos, revised through mobile MP4 reviews and delivered in two export versions.
Reported by Aaron MakelkyHow the agent did it
- Supply the source footage, supporting photos and brand-asset location.
- Transcribe and assemble the strongest spoken sections in HyperFrames.
- Review small playable MP4 copies on the phone and mark faulty frames.
Where a person stepped in
Makelky reviews multiple versions, directs corrections, supplies licensed audio after a service failure and personally publishes the finished video.
Aaron Makelky directs Codex from his iPhone to edit a fishing video using a Google Drive clip, ten photos and a screenshot of an earlier fishing report. Codex retrieves his existing brand assets, transcribes the footage and assembles a vertical HyperFrames project.
The first versions need corrections. A text panel covers too much of his face, a highlight obscures the report and the edit contains dead air. Mobile-compatible MP4 reviews let Makelky pause on an error and return a marked screenshot. Codex inspects the exact frame, uses OCR to locate the fly’s name and adjusts the highlight to the rendered words.
A generated-music service fails because the account lacks available usage. Makelky supplies a licensed track instead. The final edit runs 31.7 seconds; he publishes it on social platforms and requests a second export without music or captions for native additions.
The outcome is a published video with an inspectable revision loop, not a single successful prompt. Makelky supplies the story, real evidence, creative decisions and licensed audio. A localhost preview initially fails his remote review needs, so sharing playable files becomes part of the workflow. The useful detail is keeping the master and small review renders separate while making frame-specific corrections to one reusable project.
Prompt
Original prompt
Source: Blog · Aaron Makelky
Frequently asked questions
Why change the preview format?
A localhost preview was inaccessible from the author’s mobile session, so Codex produced playable H.264 MP4 review files.
How was the highlight repaired?
Codex inspected the exact frame and used OCR to locate the fly’s name rather than relying on a guessed rectangle.
What happened when music generation failed?
The creator supplied a track from his paid licensed-music account; the final project also had a no-music export.
More details
Remaining steps
- Inspect exact rendered frames and correct highlight placement and timing.
- Render the master and a separate native-social version without music or captions.
About this account
Codex retrieves media and brand context, transcribes, edits, renders review copies, corrects inspected frames and creates separate delivery versions.
Firsthand account by Aaron Makelky; original publication date verified against Primary field-note byline July 25, 2026. Outcomes are the author’s reports, not independently reproduced. Original 25-word mobile-review correction from the executed workflow; the full initial editing prompt is not reproduced.
Credit: Aaron Makelky
Source published 25 Jul 2026
Source checked 14 Oct 2026
Tools and topics
Codex · ChatGPT mobile app · Google Drive · HyperFrames · FFmpeg · OCR · Licensed music
Mobile video review · Frame-specific correction · Separate social exports
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.