Editing Harness

Best for: high-precision checks where edits must be exact and non-target content should remain unchanged.

This harness runs two example cases:

• vto_jacket_tryon: virtual try-on using reference person + garment
• logo_year_edit: precision logo text edit

Run

python editing_harness/run_imagegen_evals.py

Run a single case (use --cases):

python editing_harness/run_imagegen_evals.py --cases vto_jacket_tryon

Run multiple cases (comma-separated, no spaces):

python editing_harness/run_imagegen_evals.py --cases vto_jacket_tryon,logo_year_edit

Flags (when to use them)

• --cases: limit runs to specific case ids. Valid values:
  • vto_jacket_tryon
  • logo_year_edit
• --model: image model under test (defaults to gpt-image-1.5).
• --judge-model: LLM used to grade outputs (defaults to gpt-5.2).
• --num-images: number of edited images to generate per case (defaults to 1).

Inputs

This harness expects the following reference images in images/:

• images/base_woman.png
• images/jacket.png
• images/logo_generation_1.png

Outputs

Results are written under editing_harness/results/<run_id>/:

• results.json and results.csv with per-example scores
• summary.json with aggregated metrics
• artifacts/ for edited images

Adapting the harness

• Replace the input images and edit prompts in run_imagegen_evals.py.
• Update the schemas and verdict rules to fit your workflow.