CAE Capture - Clean Render Output

Capture production-quality images and movies from Kit-CAE scenes — rendered content only, no application UI.

Purpose

Capture clean render-only output from a Kit-CAE scene — high-resolution PNG/EXR screenshots and MP4 video with NVENC hardware encoding, with no application UI in the frame.

Prerequisites

cae-core (loaded automatically). Kit must be launched with the Farm-Style args from cae-core/references/capture-api.md for any capture work; mid-session carb.settings changes will crash the renderer.

Instructions

1. Select an approach — choose single image, MP4 video, or animated orbit per ## Approach Selection.
2. Launch Kit with Farm-Style args — required for any capture work; see ## Launch (MANDATORY) and cae-core/references/capture-api.md. Do not change renderer settings mid-session.
3. Frame the camera — apply the settle-frame, target-prim, and FOV tips in ## Camera Tips.
4. Capture the output — for animation follow ## Time-Varying Data Capture; for single-image use the patterns linked in ## Approach Selection.
5. Verify delivery — confirm output paths, formats, and resolution against ## Delivery.

Examples

The ## Time-Varying Data Capture and ## Camera Tips sections include runnable Kit launch commands and Python snippets covering single-image, EXR, MP4-via-NVENC, and orbit-animation workflows.

Limitations

NVENC encoding requires a supported NVIDIA GPU. Headless capture requires a virtual display. Capture is renderer-only — UI elements (gizmos, viewport overlays) are deliberately excluded by the Farm-Style launch path.

Dependencies

• cae-core/SKILL.md — Preflight, Z-up, launch args, clean exit
• cae-core/references/capture-api.md — Full capture API reference (all details)
• cae-core/references/kit-cae-api.md — Import, viz setup, framing

Always run the preflight checklist from cae-core/SKILL.md first.

Approach Selection

Goal	Approach	Reference
Single screenshot	capture_viewport_to_file	capture-api.md § Single-Frame
MP4 video (turntable, animation)	CaptureExtension	capture-api.md § MP4 Video
Per-frame control (fallback)	Manual capture + H.264 encoder	capture-api.md § Manual Frame Capture

Always try CaptureExtension first for video. Fall back to manual capture only if CaptureExtension produces incorrect frames after tuning settle latency.

Launch (MANDATORY)

Use farm-style launch args from cae-core/SKILL.md § Render-Product Capture. For volumes, add the RTX renderer args. Settings that affect the renderer MUST be launch args, never carb.settings mid-render (ERROR_DEVICE_LOST).

Time-Varying Data Capture

Pre-Cache Time Steps (ESSENTIAL for smooth video)

for frame in range(0, TOTAL_FRAMES + 1, 2):
    timeline.set_current_time(frame / FPS)
    for _ in range(20):
        await app.next_update_async()
for frame in range(1, TOTAL_FRAMES, 2):
    timeline.set_current_time(frame / FPS)
    for _ in range(10):
        await app.next_update_async()

Fixed Color Range

Lock scalar color range to prevent per-frame auto-rescaling. See cae-visualization/SKILL.md § Fixed Color Range.

Camera Tips

• Always AddTransformOp() with full matrix — Euler angles break in Z-up
• Create NEW camera prim — /OmniverseKit_Persp ignores keyframes
• Orbit in X-Y plane, elevate along Z (Z-up coordinate system)
• N+1 keyframes for seamless loop — frame N == frame 0, capture [0, N)
• end_frame is exclusive — set end_frame = N for N frames, not N-1

Full camera setup, orbit patterns, and non-constant paths: capture-api.md § Camera Animation.

Delivery

macOS / Apple Devices

Kit's NVENC encoder writes fragmented MP4 that macOS reads as 2× duration. Always remux before delivery:

ffmpeg -y -i raw.mp4 -c copy -movflags +faststart -f mp4 final.mp4

Details: capture-api.md § macOS AVFoundation Duration Quirk.

H.264 Required for Web/Chat

MP4v works locally but won't play inline on Slack/Discord/web. Use Kit's NVENC encoder (H.264 by default) or ffmpeg with -c:v libx264.

Troubleshooting

Problem	Fix
ERROR_DEVICE_LOST	Use farm-style launch args (not carb.settings)
Black/empty output	Wait ≥600 frames for convergence
UI in output	Use capture_viewport_to_file, not swapchain capture
Camera won't animate	Create new camera prim (not /OmniverseKit_Persp)
Resolution ignored (headless)	Set fillViewport=false (launch arg + carb setting)
Color shifts per frame	Disable RescaleRange, set explicit domain
Uniform color in video	Increase settle latency or use manual capture
0/N frames found, but PNGs appear after exit	Scan frame files after a final 120-300 update flush; don't rely on immediate per-frame existence checks
Frame count off by one	end_frame is exclusive — use N not N-1
Kit process survives	Add os._exit(0) after post_quit()
macOS 2× duration	Remux with ffmpeg -c copy -movflags +faststart