Skip to content

Action 6 original media

Loaner survey on 2026-09-21, Action 6 firmware V01.02.0521, Mimo 2.12.0, base lens. This verifies the files behind the settings, mode and follow-up observations. All media and unredacted metadata remain private under captures/action6-20260921/analysis/.

The operator selected all 12 survey items in Mimo Album and began batch download at 10:38:49 UTC, screenshot 414. Before download, Batch Download Settings → Custom-Cropped Files was changed from 4:3 to Original (screenshots 411–413); Photo Frame was off. The USB Photos inventory grew from 980 to 991 entries, exactly nine MP4 and two JPEG exports. The twelfth survey item, the earlier 4K4:3 D-Log M recording, was already preserved and already present on the phone.

Only the 11 paths absent from that frozen baseline were read through USB media AFC. Source paths/stat information, local names, byte sizes and hashes are in analysis/media-originals/usb-exports-provenance-private.json. The original baseline was not overwritten. No other phone photo/video contents were copied. New exports total 170,962,069 bytes; including the previously secured original, the 12 survey files total 226,262,189 bytes.

Every newly retrieved MP4’s primary video and available audio stream was fully decoded with FFmpeg, with exit 0 and zero error output. Both JPEGs also decode cleanly. The previously secured original passed a separate full decode earlier. SHA-256 hashes below identify exact bytes, not a transcode or preview proxy.

The T23 download trace contains nine MP4 and two JPEG GET requests, no DNG. Three complete HTTP objects could be reconstructed directly and match USB files byte-for-byte: both JPEGs and the Portrait Mode 2.7K4:3 recording. This provides independent camera-to-phone-to-Mac provenance for those files. Other network objects were not fully reconstructed; their complete USB files are authoritative. T23 rolled over early during download, at 10:40:20.281817 UTC, rather than lasting five minutes. Do not assume take numbers imply fixed-duration boundaries.

The table describes the primary recorded stream, not embedded JPEG covers or preview proxies. Dimensions are encoded pixel dimensions; portrait 4K is actually 2160×3840, without relying on a 90-degree display rotation. Duration is primary-video duration; AAC duration may differ by a small codec-frame boundary. Colour names come from the associated accepted Mimo setting/telemetry, because standard video colour metadata does not identify D-Log M reliably.

Local evidence file Recording context Dimensions Codec / pixel format Exact file frame rate Frames Duration, s Audio
survey-export-01.mp4 Slow Motion 1080P 8×,Normal 1920×1080 HEVC Main 10 / yuv420p10le 30000/1001 752 25.091733 No audio track
survey-export-02.mp4 Timelapse 4K4:3 30 3840×2880 HEVC Main / yuv420p 30000/1001 33 1.101100 No audio track
survey-export-03.mp4 SuperNight 4K60 3840×2160 HEVC Main 10 / yuv420p10le 60000/1001 246 4.104100 AAC LC,48kHz,stereo
survey-export-06.mp4 Video portrait 4K120,D-Log M 2160×3840 HEVC Main 10 / yuv420p10le 120000/1001 225 1.876875 AAC LC,48kHz,stereo
survey-export-07.mp4 Video 8K24,Normal 7680×4320 HEVC Main 10 / yuv420p10le 24000/1001 46 1.918583 AAC LC,48kHz,stereo
survey-export-08.mp4 Portrait Mode 2.7K4:3 24 2688×2016 HEVC Main 10 / yuv420p10le 24000/1001 65 2.711042 AAC LC,48kHz,stereo
survey-export-09.mp4 Hyperlapse 4K25,2×,D-Log M 3840×2160 HEVC Main 10 / yuv420p10le 25/1 59 2.360000 AAC LC,48kHz,stereo
survey-export-10.mp4 Custom4K 1:1 60,D-Log M 3840×3840 HEVC Main 10 / yuv420p10le 60000/1001 189 3.153150 AAC LC,48kHz,stereo
survey-export-11.mp4 Video 8K24,D-Log M 7680×4320 HEVC Main 10 / yuv420p10le 24000/1001 44 1.835167 AAC LC,48kHz,stereo

The previously secured settings-survey/USB-original-DLogM-4K43-24.mp4 is 3840×2880, HEVC Main 10 / yuv420p10le, 24000/1001 fps, 275 frames, 11.469792 s, with AAC 48 kHz stereo. It is 55,300,120 bytes and has SHA-256 ca0a9f3cafb3e4e7c662aa1f21c5733690e0e0b4aeec00a52328669cb192b991. Its detailed transfer provenance is in the settings findings.

Custom4K Original export is truly 3840×3840. Its embedded cover is 720×720, which is a separate attached picture. The recorded HEVC stream retains the full 1:1 area; the earlier default 4:3 batch-export preference did not crop this file because Original was selected before download. This validates the actual original dimensions for the sample, not every future Mimo export preference.

Both 8K samples are 7680×4320,HEVC Main 10,4:2:0,10-bit, encoded at 24000/1001 fps. The D-Log M sample is survey-export-11.mp4, corresponding to record start 10:31:52.639223, T21 frame 41012. The Normal sample is survey-export-07.mp4, after accepted Normal colour SET at 10:32:42.116699, frame 51509; recording starts 10:32:48.180703, frame 52975. Both are short samples under two seconds, so they do not qualify sustained 8K operation or thermal limits.

The Video portrait 4K120 original uses 120000/1001 fps, or approximately 119.88, whereas Hyperlapse 25 uses 25/1 exactly. Preserve rational frame rates in media metadata instead of converting every UI rate to an integer or to an NTSC fraction.

Slow Motion and Timelapse differ materially

Section titled “Slow Motion and Timelapse differ materially”

The 1080P8× Slow Motion file is 10-bit HEVC, 752 frames at 30000/1001 fps, giving 25.091733 s of playback. It has no audio track. The UI/command setting was 240-fps acquisition with 8× slow motion; the saved file’s playback rate is about 30, not 240. Multiplying playback rate by 8 gives 239.76 fps, consistent with that setting, but this is an inference from the recorded ratio and file timing, not an independent sensor-clock measurement. The file time base is 1/240000, which also must not be mistaken for its playback frame rate.

The 4K4:3 Timelapse file is HEVC Main,8-bit yuv420p, 33 frames at 30000/1001 fps, 1.101100 s, with no audio track. This is a verified exception to any blanket claim that every Action 6 recording mode is 10-bit. The selected capture interval was 0.5 s and unlimited duration; Video save format was restored before recording. The file therefore does not test JPEG+Video or Raw+Video sidecar creation. Its 33 frames do not establish exact wall-clock recording duration from the UI button interval.

The 2× Hyperlapse file contains 59 frames at 25 fps with a 2.36 s primary-video track, and does contain an AAC stereo track. Track presence and successful decoding do not establish how Mimo/camera retimes or processes Hyperlapse audio.

A thirteenth survey item was created during the attempted countdown. It was downloaded at11:14:13 UTC and retained as survey-export-13.jpg: 7168×5376,8-bit sRGB, 1,646,592 bytes, EXIF f/3,1/25s,ISO800. SHA-256: 28ff2af3c4888b943736e503306a8c65441850e5e4cd470065d77afe1334f81f. Full decode passed. The complete T30 HTTP response matches the USB file exactly, and its size matches the earlier media notification. The verified JPEG/MP4 set now contains 13 files,227,908,781 bytes (ten MP4 and three JPEG).

File Selected camera mode Dimensions Format EXIF aperture / shutter / ISO Byte identity
survey-export-04.jpg Photo M16:9,JPEG+RAW 3952×2224 JPEG 8-bit,sRGB f/3.0,1/25,ISO 981 Matches complete camera HTTP object
survey-export-05.jpg Photo L4:3,JPEG+RAW 7168×5376 JPEG 8-bit,sRGB f/3.0,1/25,ISO 1600 Matches complete camera HTTP object

Both report EXIF make DJI, model AC006, software 10.00.15.17. These EXIF strings are distinct from the camera UI firmware V01.02.0521 and network model token ac206; they must not be treated as the same version/identifier namespace. Unique identifiers and complete EXIF metadata stay private.

No DNG companion was exported by this batch. The DCIM inventory contains only the two new JPEGs, and the observed HTTP requests explicitly request JPEGs. A read-only metadata inventory of Mimo’s accessible Documents media directories found no retained media originals there either. The two JPEGs are byte-identical to camera JPEG responses, so this is not evidence of Mimo converting downloaded DNGs into JPEG during export.

All three DNG companions were subsequently retrieved and validated using read-only HTTP range requests from the connected iPhone while Mimo remained in the foreground. The Mac retained its original internet connection. The camera accepted the same media path with a lowercase .dng extension; this was tested for these three files, not established as a universal naming rule.

Local evidence file RAW CFA dimensions Bytes EXIF aperture / shutter / ISO Declared WhiteLevel
survey-photo-m169-complete.dng 3952×2224 18,391,824 f/3, 1/25 s, ISO 981 16,383
survey-photo-l43-complete.dng 7168×5376 77,875,472 f/3, 1/25 s, ISO 1600 65,535
survey-photo-13-complete.dng 7168×5376 77,846,800 f/3, 1/25 s, ISO 800 65,535

All three are little-endian DNG 1.4, backward version 1.3, with uncompressed 16-bit unsigned CFA samples, a 2×2 BGGR Bayer pattern, and BlackLevel 0. The full active area and default crop match the dimensions above. Stored sample width and declared white level describe file representation; they do not measure sensor precision or dynamic range. Preserve the different M/L white levels when normalizing RAW samples.

The full RAW is a CFA SubIFD, not the root thumbnail. The M file contains 160×90 and 1280×720 embedded previews; each L file contains 160×120 and 960×720 previews. All nine image arrays decode completely, and all 15 structural IFDs have bounded tables and tag data. Every RAW strip ends exactly at EOF. Decoding the stored samples does not validate every DJI correction opcode or the complete color-managed development pipeline. ExifTool’s metadata warnings were retained privately; no metadata repair was applied.

The transfer checked HTTP 206, Content-Range, Content-Length, stable ETag, and received body length before appending each chunk. Independent verification of all 168 chunk hashes confirmed continuous coverage without gaps or overlaps. All 168 corresponding HTTP 206 response headers were also retained in T39 and matched to the transfer records. A separate 1,024-byte probe is excluded from the complete-file totals. The three DNGs total 174,114,096 bytes. Together with the JPEG/MP4 originals, the verified collection contains 16 complete files, 402,022,877 bytes, from 13 capture items: ten videos and three JPEG+RAW photos.

Two earlier Safari copies of the M DNG were truncated by 66,171 and 14,043 bytes. Both decoded their previews but failed full RAW decoding. Their entire contents match the complete file’s prefix. The final file was downloaded anew; no padding or synthetic repair was used. A visible preview or completed download indicator is insufficient evidence of RAW completeness.

DNG SHA-256
M 16:9 583fb838c6d7dcee5c4b30e6f88e934a0e12461094363b9dd4414f497d966744
First L 4:3 5ec58eb554d13d5de66a8cb336c31150409bb40ba724e2d2978d22a5bd8e9a7f
Photo 13 eae135614b645eab5d7d125dbe458c6b71d5f9d368ceede75dd8b2eafdbd1bf3

Complete metadata includes private identity and location fields and remains outside the public documentation. A zero SubjectDistance tag is not a measured focus distance. Camera EXIF wall-clock timestamps also require the documented clock correction before correlation with UTC capture events.

Metadata interpretation and future implementation

Section titled “Metadata interpretation and future implementation”

All nine new primary video streams carry hvc1, TV/limited range and BT.709 primaries/transfer/matrix metadata, including those recorded in D-Log M. Thus a BT.709 transfer tag alone does not identify Normal versus D-Log M in these files. Retain camera colour state or independently parse validated DJI metadata before choosing a display transform; do not apply a log LUT solely from Main 10.

ExifTool’s generic QuickTime BitDepth=24 field appears even for these 10-bit HEVC files. The decoded codec profile and pixel format establish the primary stream’s sample precision. Do not interpret that generic container field as 24-bit-per-channel image precision or override yuv420p10le with it.

Each MP4 also carries one attached MJPEG cover and opaque data tracks. Their smaller dimensions are thumbnails, not alternate recording resolutions. The survey retains the complete files so data-track, timestamp and sidecar decoding can continue after the loaner leaves. No DJI metadata schema is asserted from the mere presence of those tracks.

Bitrates in these short indoor samples vary with scene content and mode; they are preserved in the private ffprobe JSON but do not establish maximum bitrate, storage planning for long recordings or card-speed requirements. Audio checks prove the observed AAC 48 kHz stereo tracks decode; microphone routing, gain, wind reduction and accessory combinations require their own control evidence.

These are local evidence labels, not original camera filenames or public media links. Camera/phone paths, device IDs and complete metadata are private.

File Bytes SHA-256
survey-export-01.mp4 30946822 e701cc3ec296b93288452d84f74f1aefe93caead59b9144d6b97a3f1a9460853
survey-export-02.mp4 6207598 d6288611ee7facb101347d02bc4622d19d5dad7471e1266108aefaccb8282860
survey-export-03.mp4 26516603 dc8eb3db039c8397b7c788d7e067b36bc78a6591a3049b2c370efb052a1d8942
survey-export-04.jpg 684032 623d3b4e5644931128daf781b1582fc58cb2c18e4b58f810d14780b6f33dabdb
survey-export-05.jpg 2035712 8572fa07005c88c46eb08a427764e15ff422c802a197bfa012e746f60f4ed264
survey-export-06.mp4 19920566 2b16df53f88bc6da510bdac71a060417d6e723ef7da78092ec4eafbfaea59db3
survey-export-07.mp4 17967194 02a50d030640032de150ba6505857236b49ba7018ea4f26e0c0bd0aa81de6af6
survey-export-08.mp4 4875808 9282c3ef65d47d78aec3968642246fa9e2bc5a1ce54820d6d2a9caa44af77094
survey-export-09.mp4 12085827 619181752a8360815972054da24e32f29c8d44df002a5957147f75ad42fdc725
survey-export-10.mp4 34340065 010aae14563ba6bdfbf3cefb46f7f0be5da6ff106c0d36ad9624d604c9473346
survey-export-11.mp4 15381842 d10bef9524f1fa77c965e4b5520ba0b925968407ab3924b2e2f91f00b2b33b2f

Verification used FFprobe stream/format inspection, ExifTool metadata inspection, SHA-256 and full FFmpeg decode of each primary video plus available audio, or JPEG image. In the initial USB-export verification, all 11 newly exported files passed with zero decoder errors. That offline validation read existing sources without camera/app interaction. The later thirteenth JPEG and three DNGs were transferred and verified separately as described above. No camera media was deleted during any of these operations.