Image Tools Plugin

mere-image-tools contains local image-production helpers for mere.run workflows.

The first supported workflow is subject knockout through the native mere.run vision segment SAM 3.1 command.

Why This Is A Plugin

Subject knockout needs matte post-processing and recipe-specific defaults that are useful in production image workflows. Keeping that workflow in a companion plugin lets the core mere.run CLI stay focused on model inference while still giving production users a stable command surface built on top of the native CLI.

Commands

mere-image-tools doctor
mere-image-tools plan \
  --input ./frame.png \
  --output ./subject.png \
  --mask-output ./subject-mask.png
mere-image-tools run ./subject.run.json
mere-image-tools knockout \
  --input ./frame.png \
  --output ./subject.png \
  --mask-output ./subject-mask.png \
  --prompt "subject"
mere-image-tools cleanup ./subject.run.json

knockout is a convenience command that writes a run manifest and executes it immediately. Use plan plus run when you want to inspect or store the operation before execution.

mere.run Runtime

The plugin calls:

mere.run vision segment ./frame.png \
  --model vision-segment-sam31 \
  --prompt "subject" \
  --output ./subject.sam31/segmented.png \
  --json-output ./subject.sam31/segmented.json \
  --mask-output-dir ./subject.sam31/masks

Override the executable with MERE_IMAGE_TOOLS_MERE_RUN or --mere-run-command when you need to target a source checkout or a non-standard install path.

Prompted subject extraction is the default because it gives SAM 3.1 the clearest target. Box and point prompts are also passed through with --box and --point. When you pass multiple --prompt values, the plugin combines the best SAM mask for each prompted label so subject-plus-prop cutouts can stay together.

The plugin owns planning, manifests, stdout/stderr discipline, mask selection, matte cleanup, transparent PNG composition, and artifact hashing. mere.run owns SAM 3.1 model resolution and segmentation.

Outputs

The plugin records:

• transparent PNG output
• grayscale mask output
• mere.run vision segment JSON and mask artifacts
• SHA-256 hashes after success
• no-op cleanup status, because no remote resources are created