Edge Case Synthetic Training Image Prompt Engine

Role: Lead Synthetic Data and Computer Vision Engineer specializing in diffusion-based perception dataset augmentation.

Context

• Target Object Class: {{target_object_class}}
• Occlusion Scenarios: {{occlusion_scenarios}}
• Environmental & Weather Conditions: {{environmental_conditions}}
• Sensor Noise & Artifact Profiles: {{sensor_noise_profiles}}
• Camera Intrinsics & Perspectives: {{camera_intrinsics}}
• Target Batch Quota: {{batch_volume_quota}}

Task

Build a Python script that algorithmically generates a structured, non-redundant matrix of photorealistic image generation prompts designed to create synthetic training datasets for {{target_object_class}}, systematically permuting {{occlusion_scenarios}}, {{environmental_conditions}}, and {{sensor_noise_profiles}} up to {{batch_volume_quota}} iterations.

Method

1. Define structured vocabulary arrays for physics-based lighting, optical aberrations, sensor artifacts from {{sensor_noise_profiles}}, and lens metrics from {{camera_intrinsics}}.
2. Implement a combinatorial permutation engine that samples distinct pairings of {{environmental_conditions}} and {{occlusion_scenarios}} without duplicate tuples.
3. Construct an explicit prompt assembler that positions {{target_object_class}} in realistic contextual grounding, enforcing physical contact with surfaces.
4. Embed precise photorealistic camera parameters (e.g., focal length, shutter angle, ISO grain, depth of field) into every prompt string.
5. Build a validation layer that checks generated prompt diversity using token overlap metrics to prevent repetitive training samples.
6. Generate an annotation metadata sidecar dictionary for every prompt (recording planned bounding boxes, weather labels, and occlusion percentages).
7. Output the completed pipeline script with CSV and JSON Lines export capabilities, ready for ingestion by image generation worker pools.

Constraints

• The script MUST output deterministically reproducible prompt batches using an explicit random seed parameter.
• The generated prompts MUST NOT contain artistic stylization buzzwords like 'hyperrealistic', 'trending on artstation', or 'masterpiece'.
• Prompts MUST enforce raw photographic attributes (e.g., RAW photo, 35mm sensor capture, natural chromatic aberration).
• The script MUST export metadata JSON alongside each prompt containing the ground-truth variable permutation state.

Output format

Provide a self-contained Python script including:

1. Combinatorial Generation Engine with seeding logic.
2. Prompt Formatting and Camera Lens Simulation Module.
3. Diversity Assertion and Output Validator.
4. CLI Runner printing batch statistics and sample JSONL records.

Self-review

• Confirm that every permutation explicitly targets {{target_object_class}} across varying {{environmental_conditions}}.
• Ensure lens and sensor attributes from {{camera_intrinsics}} and {{sensor_noise_profiles}} appear as photographic syntax rather than generic keywords.
• Check that batch generation scales gracefully up to {{batch_volume_quota}} records.