What the video workflow actually does.
A live synthetic test, downloaded-file measurements, network observations, and static analysis of the JavaScript shipped by Camuflage Ads.
Observed test completedBrowser rendering, cloud history
The React interface selects files, checks the account plan, and runs a browser media processor. Its shipped code uses canvas and WebCodecs with MediaRecorder and FFmpeg/WASM fallback paths. IndexedDB stores input Files, result Blobs, and job state. The completed output is uploaded to Supabase Storage and listed in cloud history.
The observed run requested plan validation, the renderer and fallback scripts, the base-clip catalogue, and the output-storage/history APIs. The result upload returned HTTP 200 and its history insertion returned 201. No account credentials are included in this report.
The six-second test
One synthetic portrait clip and a geometric cover were processed using UI V2, Standard, with the original-audio switch enabled and filename randomization disabled.
| Measured property | Source | Downloaded result |
|---|---|---|
| Duration | 6.000000 seconds | 6.292333 seconds |
| File size | 791,184 bytes | 913,834 bytes |
| Video | H.264 / 360 × 640 | H.264 / 360 × 640 |
| Audio | AAC / 44.1 kHz / mono | Opus / 48 kHz / stereo |
| First frame | Color test pattern | Supplied cover image |
| Extended tail | None | Not observed |
The audio changed codec and channel layout. The output also stayed at the source dimensions instead of the newer wrapper’s requested 1080 × 1920. Alongside the fallback script request, this is consistent with a fallback path, but it does not prove which internal branch failed.
What the source code establishes
- Two version names: UI V1 selects internal
v2; UI V2 selectsv4, the initial default. - Three levels: Basic, Standard, and Maximum map to
basic,normal, andaggressive. - Sequential jobs: local jobs update progress, support cancellation, and may retry before saving their result.
- Different tail paths: the older renderer can add an extended still-image tail. The newer wrapper disables that tail and can append a full randomly selected base clip. It can return the earlier result when the extra stage is unavailable. The test produced no extended tail.
- Disconnected advanced state: the inspected audio/rename switches are bound to UI state but are not passed to the processing call. The live audio comparison independently shows that the enabled switch did not preserve the encoded audio track.
- Cloud history: results are uploaded to
camouflage-results, listed throughcamouflage_history, and downloaded with signed URLs or a locally retained Blob.
These facts describe the inspected build and one synthetic run. No Meta or TikTok upload/review was performed. Platform approval and detection-related marketing claims were not verified.
What Camu Studio builds from this
This studio provides batch uploads, a cover preview and explicit visible intro, quality presets, working audio/metadata/filename controls, processing progress, cancellation, playback, downloads, and persistent history. Native FFmpeg runs on your private VPS. Settings describe actual edits to the exported file; output uses standard H.264/AAC MP4.
The implementation is original. It does not claim byte equivalence with the reference. Billing, AI cover generation, unrelated appended footage, and advertising-review behavior are outside this build.
Traceable sources
REA successfully parsed six selected feature chunks. The complete static parent Evidence is ev_29ae6a1f39f138c9371dcf11cc70115d7ab7936a8851807a7b3ccbd9b5a897a9. The separate full main-bundle analysis reached its 1.5 GiB worker heap limit, so that graph is explicitly incomplete. Downloaded-file hashes, exact source ranges, redacted network observations, and the saved REA bundle are provided with the separate evidence package.
Downloaded test output SHA-2565f42560193e41de2fd134687bfb6c3344dfe6681a63fec3f2fd7c52ed25b46ca