Image Compose plugin

mere-image-compose turns a local image generation command into a repeatable composition with recorded prompts, model settings, references, LoRAs, seed, dimensions, artifacts, and hashes.

Install and generate

mere.run plugin install mere-image-compose --yes
mere-image-compose doctor
mere-image-compose generate \
  --prompt "cinematic product render on warm paper" \
  --model image-klein-9b \
  --width 1024 \
  --height 1024 \
  --seed 42 \
  --output-dir ./image-out

Add references or a LoRA

mere-image-compose generate \
  --prompt "editorial portrait" \
  --ref-image ./reference.png \
  --strength 0.55 \
  --lora ./style.safetensors \
  --lora-scale 1.5 \
  --output-dir ./portrait-out

Planned composition

Use plan with the same generation flags to write run.json, then run it with run. This is useful when a pipeline or reviewer must approve the complete composition before inference.

The plugin delegates model behavior to mere.run image generate; it owns the repeatable plan and production record.

Image finishing graph nodes

The same installed plugin exposes four deterministic image nodes through the public graph-provider protocol. They appear in Graph Studio when the selected executor reports the mere-image-compose provider catalog.

Node	Operation
image.crop	Crop to an exact pixel rectangle.
image.mask	Apply a grayscale mask as transparency.
image.composite	Place a transparent layer over a base image with position and opacity.
image.inpaint	Fill a small masked region from neighboring pixels without a model.

All four nodes produce a PNG artifact. Inpainting is a deterministic local repair for small defects, with a 4 megapixel source limit and a mask covering at most 25% of the image. It does not use a generative image model. Images in the other operations are limited to 16 megapixels.

mere-image-compose graph catalog --json
mere-graph-conformance --provider mere-image-compose --json

To start with a connected repair and overlay workflow, export the bundled template:

mere-image-compose graph templates export image-repair-composite --output ./finishing.workflow.json

Set the source, mask, and overlay input paths in an inputs JSON document. The mask must match the source dimensions. White mask pixels mark the defect to fill. The template exposes repaired and composited outputs for comparison.