Camera & audio capture on device
Capture photos, video or audio with specific modes and settings on phones or computers (virtual scenes).
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- Camera · Voice Memos · Audio Recorder
- Class GUI-need label
G2· GUI-essential
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R12.09.01.02-T01 Timed 4:3 photo of a QR target with grid on
Original instruction
On the Android 14 emulator whose back camera shows the virtual scene 'desk_qr', open the Camera app, switch to 4:3 at the highest resolution, turn the grid on, set a 3-second timer, and take one photo in which the QR code on the desk is fully in frame and readable.
- Platforms
and· Android- Applications
- Camera
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
touch· Touch gesturesclick· Point & clickvis· Visual judgment- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Newest MediaStore image: owner_package_name = camera app, EXIF Make/Model from the emulator HAL, 4:3 aspect at max resolution; QR decodes (zbar) to the scene payload; Camera prefs show grid on and timer 3 s. No adb shell for the agent; files pushed via adb or created by other UIDs are detected -> 0.
Proposed partial credit
30% 4:3 max resolution; 20% grid and timer set; 40% readable QR in frame; 10% captured by Camera app
R12.09.01.02-T02 15-second 1080p30 clip with stabilization off
Original instruction
On the Android emulator Camera app, switch to Video, set 1080p at 30 fps, turn video stabilization off, and record a clip of 15 +/- 1 seconds of the virtual scene 'street_cam'. Keep only that one new clip.
- Platforms
and· Android- Applications
- Camera
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesclick· Point & clickrt· Real-time control- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Exactly one new MediaStore video by the camera app: ffprobe 1920x1080 (or 1080x1920), 30 fps, duration 14-16 s; Camera prefs stabilization off. No adb shell; MediaStore entries from other packages or adb pushes are detected -> 0.
Proposed partial credit
30% resolution and frame rate; 40% duration window; 20% stabilization off; 10% single clip kept
R12.09.01.02-T03 Titled voice recording with location off
Original instruction
In the Pixel Recorder app on the Android emulator (virtual mic plays fixtures/standup.wav), record 30 +/- 2 seconds, title it 'Standup 2026-10-14', and make sure location tagging is off for the recording.
- Platforms
and· Android- Applications
- Audio Recorder
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturestype· Text & formsrt· Real-time control- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Recorder DB/file: one new recording 28-32 s titled exactly, location field empty and setting off; audio cross-correlates with standup.wav (peak > 0.6). No adb shell; files from other UIDs or adb push are detected -> 0.
Proposed partial credit
40% duration and source audio; 35% title exact; 25% location off
R12.09.01.02-T04 Record, trim and enhance a Voice Memo on macOS
Original instruction
On macOS, the virtual input device 'Loopback Mic' plays fixtures/lecture_clip.wav. In Voice Memos record about 25 s, trim the memo to keep exactly 5.0 s to 15.0 s of the recording, turn on Enhance Recording, and rename it 'Lecture excerpt'.
- Platforms
mac· macOS- Applications
- Voice Memos
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & dropav· Audio/video over time- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Voice Memos CloudRecordings.db row titled exactly with duration 10.0 +/-0.15 s; audio aligns with the source segment 5-15 s of the recording (cross-correlation offset); enhanced flag set. Terminal withheld; eslogger detects afrecord/ffmpeg/sox exec or writes to the Recordings folder by non-Voice-Memos processes -> 0.
Proposed partial credit
20% recorded from right input; 45% trim exact; 15% enhance on; 20% renamed
R12.09.01.02-T05 Windows Camera video at 1080p with framing grid
Original instruction
On Windows 11 with the virtual webcam showing scene 'whiteboard', open the Camera app, set video quality to 1080p 16:9 30 fps, enable the framing grid (rule of thirds), and record a 10 +/- 1 s video. Then take one photo at the highest 16:9 resolution.
- Platforms
win· Windows- Applications
- Camera
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickrt· Real-time control- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Pictures\Camera Roll: one new MP4 1920x1080 30 fps 9-11 s and one JPEG at max 16:9; Camera app settings (LocalState) show grid on. Files must be written by WindowsCamera.exe (Sysmon file-create provenance); ffmpeg/other capture tools detected -> 0.
Proposed partial credit
35% video spec and duration; 25% photo spec; 20% grid enabled; 20% captured by Camera app
R12.09.01.02-T06 Manual exposure RAW+JPEG with Open Camera
Original instruction
On the Android emulator with the Open Camera app and virtual scene 'color_chart', set ISO 400, shutter 1/125 s, white balance 5500 K, focus mode manual at infinity, enable RAW (DNG) + JPEG, and take one photo of the chart filling at least half the frame.
- Platforms
and· Android- Applications
- Open Camera
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
touch· Touch gesturesdrag· Drag & dropvis· Visual judgment- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
DCIM/OpenCamera: one DNG and one JPEG with EXIF ISO 400, ExposureTime 1/125, WB manual; chart bbox (detector) >= 50 pct of frame; Open Camera prefs XML match. No adb shell; files not created by Open Camera's UID are detected -> 0.
Proposed partial credit
35% exposure settings in EXIF; 20% white balance; 25% RAW+JPEG pair; 20% chart framing
R12.09.01.02-T07 Photo Booth four-up with an effect on macOS
Original instruction
On macOS with the virtual webcam showing scene 'portrait_desk', use Photo Booth to take a 4-up photo with the built-in 'Sepia' effect, then export it to ~/Desktop/booth.jpg.
- Platforms
mac· macOS- Applications
- Photo Booth
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Photo Booth Library has one new 4-up item; booth.jpg is a 2x2 grid image with sepia toning (mean chroma low, hue ~30-40 deg) and four distinct frames from the webcam scene. Terminal withheld; eslogger detects imagesnap/ffmpeg capture or file writes by non-Photo-Booth processes -> 0.
Proposed partial credit
40% 4-up captured; 30% sepia effect; 30% exported to the right path
R12.09.01.02-T08 QuickTime movie recording from a chosen external mic
Original instruction
On macOS, record a 12 +/- 1 s movie in QuickTime Player (File > New Movie Recording) using the virtual camera 'Studio Cam' and microphone 'Virtual Mic 2' (playing a 1 kHz tone), quality Maximum, and save it as ~/Movies/mic_check.mov.
- Platforms
mac· macOS- Applications
- QuickTime Player
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clickrt· Real-time control- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
mic_check.mov: video from Studio Cam (scene fingerprint), audio dominant frequency 1000 +/- 5 Hz (not the default mic's noise), duration 11-13 s, codec per Maximum (ProRes/AAC or Apple PCM). File provenance: written by QuickTime Player; eslogger detects ffmpeg/avfoundation CLI capture -> 0.
Proposed partial credit
35% correct mic; 25% correct camera; 25% duration; 15% quality setting
R12.09.01.02-T09 Interview recording with transcript corrections
Original instruction
In the Pixel Recorder app (virtual mic plays fixtures/interview.wav, 90 s), record the full interview, then edit the generated transcript to correct the 3 mistakes listed in fixtures/corrections.txt and export the transcript as text to Downloads/interview.txt.
- Platforms
and· Android- Applications
- Audio Recorder
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturestype· Text & formsav· Audio/video over time- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Recording 88-92 s correlating with the source; Downloads/interview.txt contains the 3 corrected phrases and WER vs gold transcript <= 0.10. No adb shell; files pushed or created by other UIDs are detected -> 0.
Proposed partial credit
30% full recording; 40% three corrections; 30% exported transcript quality
R12.09.01.02-T10 Slow-motion capture of a falling object
Original instruction
With Open Camera on the Android emulator (scene 'drop_test' plays a ball drop every 6 s), record a slow-motion video at the highest high-speed frame rate available, covering at least one complete drop from release to first bounce.
- Platforms
and· Android- Applications
- Open Camera
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
touch· Touch gesturesrt· Real-time controlav· Audio/video over time- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
New video by Open Camera's UID: capture frame rate metadata >= 120 fps; grader detects the ball and finds a release and first bounce within the clip. No adb shell; files from other UIDs or adb push detected -> 0.
Proposed partial credit
40% high-speed mode used; 45% complete drop captured; 15% single file
R12.09.01.02-T11 Scan a 3-page document with perspective correction
Original instruction
On the Android emulator, use the Google Drive document scanner on the virtual scene 'tilted_doc' (3 pages shown in sequence when you tap 'next page' in the scene control) to create one PDF named 'Lease scan' in Drive (test account), with all 3 pages cropped to the page edges.
- Platforms
and· Android- Applications
- Google Drive scanner
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesdrag· Drag & dropvis· Visual judgment- Work product
WP09· Files & folders- Conditions
multiapp· Multi-app- Proposed difficulty
- 4 / 5
Proposed verifier
Grader fetches the PDF from the test Drive with its own credentials: 3 pages; each page's quad rectified (page edge straightness, aspect within 3 pct of A4); OCR CER <= 5 pct vs source text. Agent has no Drive API token; uploads by non-scanner clients are detected via Drive activity -> 0.
Proposed partial credit
30% three pages captured; 35% perspective-corrected crops; 25% OCR quality; 10% file name
R12.09.01.02-T12 Tap-to-focus and lock on a distant target
Original instruction
In the Android Camera app with virtual scene 'depth_test' (sign far, plant near), tap to focus on the far sign, lock AE/AF, recompose so the plant is centered, and take a photo where the sign is sharp and the plant is soft.
- Platforms
and· Android- Applications
- Camera
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
touch· Touch gesturesvis· Visual judgmentgeom· Precise geometry- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Grader computes Laplacian variance: sign region sharpness >= 2x plant region; plant bbox center within 10 pct of frame center; EXIF focus distance consistent with the far plane. No adb shell; media from other UIDs detected -> 0.
Proposed partial credit
50% far target in focus; 30% recomposed framing; 20% captured by Camera app
R12.09.01.02-T13 Capture the moment a signal light turns green
Original instruction
The virtual scene 'traffic_light' switches from red to green once at a random time 3-8 s after you start. Using the Android Camera app in photo mode, take a photo that shows the light green, with less than 300 ms delay after the switch.
- Platforms
and· Android- Applications
- Camera
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesrt· Real-time controlvis· Visual judgment- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Scene log records the switch time; photo EXIF/sensor timestamp minus switch time in 0-300 ms; color classifier confirms green. Exactly one photo allowed (burst or multiple shots -> 0). No adb shell; files from other UIDs detected -> 0.
Proposed partial credit
40% green captured; 45% latency under threshold; 15% single shot
R12.09.01.02-T14 Lossless 48 kHz stereo recording with safe peaks
Original instruction
In Easy Voice Recorder on the Android emulator, set format WAV, 48 kHz, stereo, and adjust the gain so the virtual mic's test song (fixtures/song.wav) peaks between -6 and -1 dBFS without clipping. Record 20 s and name it 'gain_check'.
- Platforms
and· Android- Applications
- Easy Voice Recorder
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesdrag· Drag & dropav· Audio/video over time- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
File gain_check.wav: 48000 Hz, 2 channels, PCM; true peak in [-6, -1] dBFS, 0 clipped samples; 19-21 s; correlates with song.wav. No adb shell; files created by other UIDs are detected -> 0.
Proposed partial credit
35% format settings; 45% gain within peak window; 20% duration and name
R12.09.01.02-T15 iPhone camera formats and 4K60 clip
Original instruction
On the lab iPhone (iOS 18, test Apple ID, virtual scene projected), set Camera Formats to 'Most Compatible', Record Video to 4K at 60 fps, turn on 'Lock White Balance', and record a 5 +/- 0.5 s clip of the scene.
- Platforms
ios· iOS/iPadOS- Applications
- Camera · Settings
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gestures- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Grader pulls the newest asset via the device lab's Photos export: H.264 (not HEVC), 3840x2160, 60 fps, 4.5-5.5 s; Settings plist values for formats, video, white-balance lock. Asset creator bundle must be com.apple.camera (imported files detected); the agent has no device shell.
Proposed partial credit
30% Most Compatible format; 30% 4K60 setting; 15% white balance lock; 25% clip spec
R12.09.01.02-T16 Track a moving subject with 2x zoom
Original instruction
In the Android Camera app, scene 'runner' shows a person crossing the room for 10 s. At 2x zoom, record the whole crossing while panning (by dragging the virtual camera joystick) to keep the runner in frame.
- Platforms
and· Android- Applications
- Camera
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
touch· Touch gesturesdrag· Drag & droprt· Real-time control- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 5 / 5
Proposed verifier
Video: zoom ratio metadata 2.0x; person detector finds the runner fully inside the frame in >= 0.95 of frames and bbox center within the middle 60 pct of width in >= 0.8 of frames; duration covers the crossing. No adb shell; media from other UIDs detected -> 0.
Proposed partial credit
20% 2x zoom; 50% runner kept in frame; 30% centered tracking
R12.09.01.02-T17 Capture on the phone and deliver to a Mac folder
Original instruction
Take a photo of the virtual scene 'receipt' with the Android Camera app, share it to the Mac via the lab's Quick Share bridge, and on macOS rename it to receipt_2026-10-14.jpg in ~/Documents/Receipts/ without re-encoding.
- Platforms
and· Androidmac· macOS- Applications
- Camera · Quick Share · Finder
- Interfaces
mob· Mobile touchdesk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesclick· Point & clickmulti· Cross-app- Work product
WP09· Files & folders- Conditions
xdevice· Cross-devicemultiapp· Multi-app- Proposed difficulty
- 3 / 5
Proposed verifier
Mac file SHA-256 equals the phone's MediaStore original (created by the camera app UID); path and name exact; receipt text legible (OCR finds the total). No adb shell; adb pull and Terminal use on the Mac are detected (adbd log, eslogger) -> 0.
Proposed partial credit
30% photo captured; 40% transferred byte-identical; 30% renamed and filed
R12.09.01.02-T18 Execute an eight-shot list with per-shot settings
Original instruction
Using Open Camera on the Android emulator and virtual scene 'product_table', capture the 8 shots in fixtures/shot_list.md (wide, medium, close-up, low angle, high angle, macro, video pan 6 s, timelapse 20 s) each with its specified aspect, zoom and mode, saving them in order.
- Platforms
and· Android- Applications
- Open Camera
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
touch· Touch gesturesdrag· Drag & dropvis· Visual judgmentrt· Real-time control- Work product
WP09· Files & folders- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Grader checks 8 files in capture order: aspect/zoom/mode metadata per shot, framing via detector (subject size class, camera height from horizon line) and durations for video/timelapse. No adb shell; files from other UIDs detected -> 0.
Proposed partial credit
35% all eight shots captured in order; 35% per-shot settings; 30% framing matches list
R12.09.01.02-T19 Replace the middle of a Voice Memo on iPhone
Original instruction
On the lab iPhone, in Voice Memos record 30 s of the room speaker playing fixtures/podcast_a.wav, then use Edit > Replace to overwrite seconds 10-20 with new audio while the speaker plays fixtures/podcast_b.wav. Name it 'Patched take'.
- Platforms
ios· iOS/iPadOS- Applications
- Voice Memos
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesdrag· Drag & dropav· Audio/video over time- Work product
WP05· Time-based media- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Grader exports the memo from the device lab: duration 30 +/- 1 s; 0-10 s and 20-30 s correlate with podcast_a, 10-20 s with podcast_b (boundaries within 0.5 s); title exact. Agent has no device shell; imported (non-recorded) memos are detected by missing recording metadata.
Proposed partial credit
25% base recording; 50% replace segment accurate; 25% title and duration
R12.09.01.02-T20 Front-camera portrait photo with mirroring off
Original instruction
In the Android Camera app, switch to the front camera (virtual scene 'selfie_sign' shows a person holding the text 'LAB 7'), turn off 'Mirror front photos' in settings, and take one portrait-orientation photo in which the sign text reads correctly.
- Platforms
and· Android- Applications
- Camera
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
touch· Touch gesturesclick· Point & click- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Newest image by the camera app: EXIF from the front camera HAL, portrait orientation; OCR reads 'LAB 7' unmirrored; settings pref mirror=false. No adb shell; files from other UIDs detected -> 0.
Proposed partial credit
30% front camera used; 40% unmirrored text; 30% portrait orientation and setting
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.