Pocket 3 shooting modes and formats
這一頁還沒有中文版,以下為英文原文。
Part of the Osmo Pocket 3 reference. Read the overview for firmware, survey scope and evidence levels.
Shooting modes and formats
Section titled “Shooting modes and formats”These are the observed Mimo choices and selections. A frame-rate label in Slow Motion describes the shooting setting; inspect the original before treating it as the encoded playback rate.
| Mode | Observed format menu | Exercised selections and evidence |
|---|---|---|
| Video, landscape | 1080P, 2.7K, 4K; 24/25/30/48/50/60 | All six rates selected at 2.7K; 1080P/60 and 4K/60 also selected. UI, accepted; status corroborates several settled combinations. |
| Video, square | 1080P (1:1), 2160P (1:1), 3K (1:1) | Each selected at 60. UI, accepted; square preview visible. 3K/60 recording start/stop accepted and recording status followed. |
| Video, portrait | 3K/25 exercised after body Lock Portrait | Operator/UI context identifies Lock Portrait and D-Log M; a fully validated 1728×3072 HEVC 10-bit original demonstrates native portrait composition. Other combinations and command mappings remain unverified. |
| Low-Light | 1080P and 4K; 24/25/30 | All six combinations selected with UI and accepted writes; separate status confirms several settled values. No 2.7K or square choices appear in this menu. |
| Slow Motion, 4K | 4X(100), 4X(120) | Both selected with UI, accepted and status evidence. The tested menu includes 100 although the general product specification lists 120. |
| Slow Motion, 2.7K | 4X(120) | Selected in UI, with an accepted request; no 100 option visible in this state. |
| Slow Motion, 1080P | 4X(120), 8X(240) | Both selected with UI, accepted and status evidence. |
| Photo | Frame 16:9 or 1:1; Countdown Off/3s/5s/7s | Both frames and all timer choices selected in UI. The timed square JPEG+RAW shot has a preserved JPEG and independently validated full-resolution DNG. |
| Panorama | 180° and 3×3 grid; Countdown Off/3s/5s/7s | Both capture sequences exercised in UI. Stitched JPEGs, 13 JPEG components and the RAW-selected grid’s nine DNG components have File evidence. |
| Timelapse | 1080P, 2.7K, 4K; 25/30 | All six combinations selected in UI. Interval, Duration and Timelapse Mode have separate controls. Short and full Raw+Video takes have matching nine/151 video frames and validated DNG counts. |
| Motionlapse | L to R, R to L, Custom Motion within Timelapse Mode | Settled menus confirm the choices; preview and a custom recording flow exercised in UI. |
| Hyperlapse | 1080P, 2.7K, 4K; 25/30 | All six combinations selected in UI; Speed offers Auto/2X/5X/10X/15X/30X. |
DJI’s published portrait Video sizes are 1080×1920, 1512×2688 and 1728×3072, each at 24/25/30/48/50/60 fps. The 1728×3072/25 original is measured below; the other published combinations remain a comparison baseline. DJI specifications.
The existing resolution/fps request is confirmed for Pocket 3 Video and Low-Light:
0x02/0x18: [resolution:u8] [fps-index:u8] 00 00 00| Resolution label | Resolution byte |
|---|---|
| 1080P landscape | 0A |
| 2.7K landscape | 2D |
| 4K landscape | 10 |
| 1080P square | 69 |
| 2160P square | 6A |
| 3K square | 6B |
Video fps indices 01–06 map to 24, 25, 30, 48, 50 and 60 respectively.
Aspect is carried by the resolution byte. Repeated requests sometimes precede
the matched successful reply; an unpaired initial transmission is not itself
a rejection. The command catalog
explains OpenPocketCine’s normal-Video fallback when this model does not supply
a usable capability table.
Slow Motion uses the same opcode with a different trailer. The following exact five-byte requests were accepted:
| Selection | Payload |
|---|---|
| 4K, 4X(100) | 10 0A 00 04 00 |
| 4K, 4X(120) | 10 07 00 04 00 |
| 2.7K, 4X(120) | 2D 07 00 04 00 |
| 1080P, 4X(120) | 0A 07 00 04 00 |
| 1080P, 8X(240) | 0A 08 00 08 00 |
Byte 3 follows the displayed 4X/8X multiplier in these samples. Its complete semantics are not established, but replacing it with the normal-Video zero trailer would not reproduce these Mimo requests.
Low-Light selection sends shooting mode 28 through 0x02/0xE1, with an
accepted reply and direct status mode 40. The core currently names that value
superNight; Pocket 3’s Mimo interface presents it as Low-Light video.
Native portrait recording
Section titled “Native portrait recording”In a later physical pass, the operator selected Lock Portrait on the body; the Mimo/operator context identified 3K/25 and D-Log M. The recorded camera original is 49,045,559 bytes, with 1728×3072 HEVC Main 10, 10-bit YUV420, 25fps, 151 video frames / 6.040s, plus AAC-LC, 48 kHz stereo audio lasting 6.016s. All 47 contiguous HTTP 206 ranges validated, the complete SHA-256 matched the transfer manifest, and the primary audio/video streams fully decoded without error. The subsequently preserved phone import also has an identical full SHA-256 and length. It is another copy of this recording, not an additional original.
The container’s primary video track has an identity rotation transform and 1728×3072 dimensions, also present in its video sample description. A decoded first frame viewed with automatic rotation disabled is upright and fills the 9:16 canvas without visible letterboxing. This establishes native portrait file composition for this take, separately from the letterboxed webcam outputs.
The D-Log M label comes from the captured UI/operator context; BT.709 metadata does not measure a log curve. This original was recorded after the earlier complete card copy and is an additional asset, not part of that inventory. No new accepted portrait opcode is asserted, and other portrait formats, color/codec combinations and orientation-lock persistence remain unverified.
Mimo’s Photo format control is an aspect selector named Frame: 16:9 displays an 8MP label, and 1:1 displays 9MP. Those rounded labels do not establish exact original dimensions. Countdown offers Off, 3s, 5s and 7s; each was selected. Photo Format offers JPEG and JPEG+RAW. A 1:1 JPEG+RAW shot at the 7s setting displayed a countdown, returned to idle and reduced the remaining-shot count. Both its JPEG and DNG were subsequently downloaded from the camera. The preserved JPEG outputs measure 3840×2160 for 16:9 and 3072×3072 for 1:1; full SHA-256 matches establish that the corresponding Mimo imports are byte-identical to those camera files.
The square shot’s 19,314,420-byte DNG contains a full-resolution 3072×3072 CFA image, with one sample per pixel, an RGGB 2×2 pattern and uncompressed 16-bit sample storage. Its 18,874,368-byte RAW array is separate from the embedded JPEG previews. Independent TIFF/IFD traversal checked referenced storage and image segments against the complete file. This establishes actual RAW content; 16-bit storage does not establish effective sensor precision. Demosaicing, image quality and dynamic range were not measured.
The Photo Pro list contains Grid, Focus Mode, White Balance, Overexposure Alert, Histogram, Photo Format and Timecode Display. It has no Color or audio rows in the observed state. The Focus popup offers Single and Continuous. Auto exposure displayed an ISO MAX endpoint of 6400, and EV adjustment reached −3.0 and +3.0. In manual exposure, all displayed ISO choices were selected and matched the HUD: 50, 100, 200, 400, 800, 1600, 3200, 6400. Shutter reached 1/8000s and 1s, with matching dial/HUD values and no further choices beyond those endpoints. These are UI results, not measured shutter timing or exposure calibration.
Accepted requests and separate exposure reports corroborate the Photo settings:
ISO 50 uses 0x02/0x2A value 02, and ISO 100–6400 use 03–09.
EV uses 0x02/0x2E; 07 corresponds to −3, 10 to zero and 19 to +3,
with the endpoints echoed at exposure-status byte 6. The 1/8000 and 1s shutter
requests (0x02/0x28) were also accepted and echoed. The 1s payload is
01 01 00 00 00 00 40, with the reciprocal bit clear. Some intermediate
reciprocal settings carry an additional fractional byte; an integer-only
denominator parser loses that information. Its units need matching UI evidence.
Photo entry used shooting mode 05, with an accepted 0x02/0xE1 request and
direct camera status corroboration. A photo-shutter request 0x02/0x01 with
payload 01 was accepted. The shared ShootingMode.photo value 17 belongs to
other model observations; do not use it as the Pocket 3 encoding.
Additional accepted, UI-correlated Photo requests were:
| Action | Opcode and observed payload |
|---|---|
| Select square frame | 0x02/0x12, 00 03 |
| Select JPEG+RAW | 0x02/0x16, 02 |
| Select countdown | 0x02/0x4A, 00 01 [seconds:u16le]; 3, 5 and 7 each accepted |
Independent cam_photo_param reports corroborate square at byte 1 (03),
JPEG+RAW at byte 3 (02), and countdown duration at byte 7 (3/5/7). Its byte 11
also descends from 7 to 0 during the timed shot. Only the observed choices are
mapped here. These reports do not prove the photo was written or measure the
timer’s physical delay; the preserved files provide the separate write evidence.
Panorama
Section titled “Panorama”Pano’s Type menu offers 180° and a 3×3 grid icon. Countdown shows the same Off/3s/5s/7s choices as Photo, though each delay was not exercised in Pano. Its settings panel is headed Photo and contains the same visible Pro controls. Its file-format options are JPEG and RAW, unlike Photo’s JPEG+RAW label.
The 180° capture sequence displayed progress then restored the idle monitor. The grid sequence displayed 5/9 during capture. Preserved downloaded JPEGs measure 4096×1536 for the 180° take and 4000×3840 for the grid take. The later card copy contains four 3072×3072 JPEG components for this 180° take and nine 3072×3072 JPEG components for this grid take. These are observed counts for the preserved takes, not a general component-count rule. Stitch quality remains unmeasured. The three stitched JPEG downloads, including the later RAW-selected take, have full SHA-256 matches to their camera HTTP files and copied card files.
A later RAW-selected 3×3 capture also showed progress 8/9 and returned to idle. Its album result displayed a rendered still and Downloaded status. The preserved Mimo download is independently identified as JPEG, 4000×3840. The later card copy also preserves nine 3072×3072 DNG components associated with this take. Every full-resolution CFA array was independently validated; this establishes the RAW source set separately from the stitched JPEG.
The captured mode entry uses 0x02/0xE1 value 0C. Type selection uses
0x02/0x6E, with 05 for 180° and 07 for the 3×3 grid. Pano shutter
uses 0x02/0x01 payload 07, distinct from ordinary Photo’s 01. These
requests were accepted in the corresponding UI sequences; output validation
remains separate. cam_pano_params byte 0 independently echoes type 05/07.
Selecting RAW uses a separate 0x02/0xE7 request, payload 01 00.
It was accepted, and cam_pano_params byte 1 changed to 01 while the type
remained 07. Only this RAW value is established; other format values are
not assigned labels here.
Timelapse
Section titled “Timelapse”The format popup offers landscape 1080P, 2.7K and 4K, each with 25 or 30 fps; all six combinations were selected. It shows no square or 24/50/60 choices in the observed state. A separate dropdown contains Duration, Interval and Timelapse Mode dials. The following choices were observed across the dial sweeps; the interval and duration inventory was collected in Fixed Angle:
| Control | Observed options |
|---|---|
| Interval | 0.5, 1, 2, 3, 4, 5, 6, 8, 10, 15, 20, 25, 30, 40, 60 seconds |
| Duration | Unlimited; 5, 10, 20, 30 minutes; 1, 2, 3, 5 hours |
| Suggested interval labels | Crowds at 0.5/1s; Clouds at 2s; Sunset at 3s |
| Timelapse Mode | Fixed Angle, L to R, R to L, Custom Motion |
At the longest interval the dial displays 1m while the summary displays 60s. With duration 5h, interval 60s and output 30 fps, Mimo estimates a 10-second result. This is an estimate in the UI, not a completed five-hour recording. The recorded five-minute result is documented below.
The Fixed Angle Timelapse Pro list contains Grid, Focus Mode, White Balance, Overexposure Alert, Histogram, Format and Timecode Display. Color and audio rows are absent from that observed panel. Format selects saved media, with choices Video, JPEG+Video and Raw+Video; it is distinct from the resolution/fps popup.
In the Raw+Video sequence, the selected interval became 2s. The interval dial showed 0.5s and 1s in red. Attempting 1s opened a warning that the interval was too short to save timelapse photos, with CANCEL and Save Video Only actions. Cancel retained 2s and Raw+Video. Red values are therefore an entry to a downgrade decision, not simply untappable options. The equivalent JPEG+Video warning flow has not been established. A short Raw+Video take entered recording and later returned to idle. Its preserved output has nine video frames, and the card copy supplies nine 3840×2160 DNGs. Their whole-second EXIF timestamps span 16 seconds with eight 2-second steps. A subsequent five-minute Raw+Video take at 2s, 4K/30 ran through the configured duration and returned to idle automatically. Camera counters reached elapsed 300 seconds and 151 frames before resetting, with no stop request at completion. The preserved downloaded video independently contains 151 frames, lasts 5.038367 seconds, and is 3840×2160 HEVC, 8-bit, at 30000/1001 fps. The card copy independently supplies 151 3840×2160 DNGs, with validated full-resolution CFA arrays. Their EXIF timestamps span 300 seconds: 148 successive differences are 2s, one is 3s and one is 1s. These timestamps have whole-second resolution, so they do not measure subsecond exposure cadence. The RAW count agrees with both the status counter and video frame count. Mimo’s displayed 30 is not an exact 30/1 in this file.
Timelapse enters mode 02 through 0x02/0xE1. Its start/stop uses
0x02/0x01 with 01/00, rather than the Video record opcode 0x02/0x02.
These observed requests were accepted. Mimo also sends a 16-byte
0x02/0x6C configuration; controlled menu changes correlate with these fields:
| Byte offset | Observed field |
|---|---|
| 0–1 | Constant 04 00 in these samples |
| 2 | Save type: 00 Video, 02 JPEG+Video, 03 Raw+Video |
| 3–4 | Interval as u16 little-endian, in tenths of a second |
| 5 onward | Duration as a little-endian value in seconds; 5min through 5h exercised |
| Remaining bytes | Unresolved; preserve them |
This is a partial mapping, not a complete configuration builder. The duration field’s full width is unproven because the tested finite durations fit in 16 bits. Unlimited duration, motion paths, and interactions with other fields need independent qualification before replay.
Separate 21-byte cam_lapse_param reports corroborate save type at byte 0,
interval at bytes 1–2 in tenths of a second, and duration beginning at byte 5.
The save-format comparison reveals a camera-side adjustment: a JPEG+Video request
at 0.5s was accepted and reported save type 02, interval 5. Switching to
Raw+Video still requested 0.5s, but the accepted write was followed by save type
03, interval 20 (2s). A second u16 field at bytes 3–4 changed from 5 to 20
as well; its possible minimum-interval meaning remains a candidate. Preserve
these reported values rather than assuming the requested interval took effect.
Motionlapse
Section titled “Motionlapse”Selecting L to R in Timelapse Mode exposed a Preview action. The format changed from 4K/30 to 2.7K/25 while interval 2s and duration 5min remained. This could be remembered mode state; it is not evidence of a resolution/rate restriction. R to L was visible as a further choice. Preview was exercised, and a banner indicated that tapping the screen stops it. Subsequent screenshots had already returned to the dropdown; they do not prove complete preset travel.
Settled menus also confirmed R to L and Custom Motion. The custom editor exposes waypoint thumbnails, add/delete controls and a preview action. Waypoints were added and preview exercised. The resulting layout is consistent with the fourth-point action, but the saved view does not independently establish four distinct simultaneous waypoints or their angles. A custom recording showed active recording with changed framing, then later returned to idle. Exact path, speed and interruption behavior remain unverified. The preserved short output’s file properties are listed in the original media reference; they do not establish the full planned path.
In Custom Motion, the Pro list showed Format Video and no Focus Mode row. The earlier Fixed Angle Timelapse panel had Focus Mode Single. Keep this mode-specific visibility separate from the general Timelapse control inventory.
Motionlapse uses shooting mode 18, distinct from Fixed Angle Timelapse’s
02. Parameter 0x0037 via 0x02/0x8E has a one-byte value selecting
Custom 00, L to R 01, R to L 02. Preview parameter 0x0036 has a
two-byte value, with accepted values 00 00 for Custom and
01 00 for L to R. A cam_motionlapse_params report’s byte 3 became 80 during
preview. These are observed sequences, not a complete preview start/stop enum.
The motion status layout has an 8-byte header followed by 12-byte point
records, with the count at header byte 7. Motion’s 0x02/0x6C configuration
is 16 bytes, starting with 05; only these point-edit selectors at byte 1 were
observed:
| Point action | Selector |
|---|---|
| Save C / delete C | 0D / 0E |
| Save D / delete D | 11 / 12 |
The coordinate bytes at offsets 9–14 and the final byte remain opaque. Do not derive selectors for other points or build a full path writer from these two pairs. Deletion was followed by changes in surviving point coordinates, so the experiment does not establish that pre-existing waypoints were restored or that point edits leave neighboring records unchanged.
Hyperlapse
Section titled “Hyperlapse”Hyperlapse is a separate mode on the rail. Its format popup offers landscape 1080P, 2.7K and 4K, each at 25 or 30 fps; all six pairs were selected. Its Speed dial offers Auto, 2X, 5X, 10X, 15X and 30X. These are UI setting labels; the preserved outputs’ frame rates are listed in the original media reference, while the actual speed ratio and Auto speed decisions remain unverified.
The observed Pro controls include Grid, Focus Mode, White Balance, Channel, Wind Noise Reduction, Directional Audio, Overexposure Alert, Histogram and Timecode Display. The list ends without a saved-media Format row. Focus offers Single and Continuous. There is no Color row between White Balance and Channel. Stereo and the microphone settings are visible, but those rows alone do not establish an audio track in a Hyperlapse original.
Starting Auto and 2X Hyperlapse takes displayed the stop button, recording timer and microphone meter; later screens confirmed return to idle. During recording a separate 1X / Auto or 1X / 2X choice appeared with the configured speed selected. The captured frames did not show a successful switch to 1X. This is a different control from the idle Speed dial; its effect on captured timing and audio remains unverified.
Hyperlapse mode selection uses 0x02/0xE1 value 0A, with an accepted
reply. Its 16-byte 0x02/0x6C configuration begins with 0B. Byte 3 took
decimal values 15, 10, 5, 2 and 0 alongside the speed sweep; these are candidate
speed values, not Timelapse interval tenths. Independent status confirmation,
the remaining fields and complete speed encoding still need qualification.
OpenPocketCine portrait format picker
Section titled “OpenPocketCine portrait format picker”A later operator report found that vertical 3K appeared as only 1080p/4K in
OpenPocketCine’s FORMAT picker. The reproduced empty-list fallback discarded
the reported current resolution. Both platforms now retain a reported size
such as 3K 9:16 when the effective format list is empty, and an fps selection
keeps that resolution byte. The empty-table Video fallback offers only the
accepted 16:9 and 1:1 0x02/0x18 pairs; catalog 9:16 bytes are not a SET
without Lock Portrait. Reported capabilities and the confirmed Pocket 3
normal-Video matrix still take precedence.
After the corrected app was installed on an iPhone 16 Pro Max on 2026-09-11, the operator confirmed that the vertical 3K picker worked. Automated tests cover the empty-list case and preserving portrait on an fps change. Physical Android and on-camera fps-change checks remain pending. The operator’s earlier session inputs were not captured, so the reproduction does not establish that this fallback caused that session’s behavior. See format fallback behavior.