Transform local media with ak:media-processing

Use ak:media-processing for local image, audio, and video transformations with FFmpeg, ImageMagick, and RMBG. The Skill covers format conversion, encoding, filters, thumbnails, batch images, background removal, HLS or DASH packaging, and explicitly authorized live streaming.

Choose deterministic media tooling

Use ak:media-processing when

• You need repeatable local conversion, resize, crop, compositing, filtering, audio extraction, thumbnail, GIF, or codec work.
• A batch needs explicit naming, output directories, success counts, and failures.
• You need HLS or DASH files from a local source.
• You need local AI background removal and accept a model download.
• You want to inspect media properties before selecting output settings.

Choose another workflow when

• You need semantic OCR, transcription, scene understanding, or provider media generation. Use ak:ai-multimodal.
• You need a curated visual prompt and new AI image. Use ak:ai-artist.
• You need an HTML scene rendered into a scripted animation. Use ak:html-video.
• You do not have permission to transform, remove metadata from, watermark, broadcast, or redistribute the source.

Prepare sources, targets, and tools

Before you start:

• Complete Onboarding, and confirm Engineer Kit is installed for the current runtime and scope.
• Install only the required local tools: FFmpeg and ffprobe for video or audio, ImageMagick for still images, and rmbg-cli for background removal. Verify with ffmpeg -version, ffprobe -version, magick -version, or rmbg --version as applicable.
• Preserve originals in a separate location and choose new output paths. Avoid an in-place mogrify command unless overwrite is intentional and recoverable.
• Inspect codecs, streams, dimensions, frame rate, duration, color space, metadata, file size, and target playback requirements before conversion.
• Estimate output count and free disk. Long encodes, high-resolution images, batch concurrency, and local models can consume substantial CPU, GPU, memory, temporary storage, and time.
• For live streaming, name the exact endpoint, account, stream key handling, start and stop window, audience, and approval to publish.

Run a reversible transformation

/ak:media-processing "Inspect ./media/demo.mov, create ./artifacts/demo-web.mp4 as H.264/AAC with fast-start and a maximum width of 1920, preserve the source, test settings on a 15-second sample first, and report size, duration, streams, and visual quality"

/ak:media-processing "Inspect ./media/demo.mov, create ./artifacts/demo-web.mp4 as H.264/AAC with fast-start and a maximum width of 1920, preserve the source, test settings on a 15-second sample first, and report size, duration, streams, and visual quality"

$ak:media-processing "Inspect ./media/demo.mov, create ./artifacts/demo-web.mp4 as H.264/AAC with fast-start and a maximum width of 1920, preserve the source, test settings on a 15-second sample first, and report size, duration, streams, and visual quality"

Match the tool to the effect

The bundled Python helpers support --dry-run; inspect the actual emitted command before a large batch. batch_resize.py also supports bounded parallel workers, while media_convert.py offers web, archive, and mobile presets.

Observe the processing stages

1. The Skill confirms ownership and outcome. It records input scope, output paths, target devices or service, quality goals, and destructive boundaries.
2. It inventories the source. ffprobe, identify, or tool-specific checks capture streams, codecs, dimensions, duration, metadata, and file size.
3. It checks local capability. Installed encoders, decoders, delegates, policies, models, hardware acceleration, and writable storage are verified.
4. It plans a sample. Codec, quality, scale, crop, filter order, audio, metadata, and naming are made explicit and tested on a small representative input.
5. It writes new outputs. The transformation runs with bounded threads or batch concurrency and preserves logs or failures needed for diagnosis.
6. It validates the files. Output probes, playback or visual inspection, dimensions, duration, stream presence, file size, transparency, and quality are compared with the contract.
7. It reports cleanup. Outputs, temporary files, model downloads, skipped items, failures, and whether sources remain untouched are summarized.

Control overwrites, publication, and local resources

• Do not pass broad globs until the matched file list and output collision behavior are reviewed. Keep batch outputs outside the input directory when possible.
• Metadata stripping can improve privacy but can also remove attribution, orientation, color profiles, timestamps, or evidence. Decide what to retain.
• Stream keys and encryption keys are secrets. Keep them out of command history, logs, reports, playlists, and source control.
• RMBG processes locally, but installation and first model use require network downloads. The bundled shell helpers can install rmbg-cli globally when it is missing; do not run them without approving that package-manager mutation.
• Hardware acceleration and high concurrency can monopolize shared GPUs, CPU, memory, and I/O. Set resource boundaries and stop when the host becomes unstable.
• Codec, container, browser, device, social-platform, and broadcast support vary by installed build and target. Verify the actual destination rather than treating a successful encode as universal compatibility.

Verify the outputs

A complete result should include:

• Source inventory, tool versions, command or helper used, and the exact target settings.
• Output paths, counts, dimensions, duration, codecs, streams, file sizes, and metadata or transparency state.
• Sample review and final visual or playback checks, including quality loss, crop, sync, color, audio, and compatibility findings.
• Source preservation, overwrite decisions, temporary-file cleanup, and failed or skipped batch items.
• CPU or GPU path, concurrency, elapsed time, model download, network endpoint, and any external publication effect.

Stable and beta package identical ak:media-processing sources and helpers. The release does not prove that every host FFmpeg build, ImageMagick delegate, RMBG download, hardware encoder, target device, or streaming service behaves alike.

Troubleshoot and interpret limits