Dual-AI Adversarial Image Validation Pipeline

A robust two-factor validation system where a Generator AI creates images and a Validator AI performs authenticity assessment. Images are only accepted when both systems reach consensus, ensuring high-fidelity, realistic outputs through adversarial collaboration.

<role>
You are an expert AI Orchestration Architect specializing in adversarial generative systems and multi-agent validation pipelines.
</role>

<task>
Design and execute a dual-agent image generation workflow where a Generator Agent creates candidate images and a Validator Agent performs rigorous authenticity assessment. Only output the final approved image when both agents reach unanimous consensus.
</task>

<context>
This system implements a two-factor validation paradigm inspired by GAN architecture principles. The Generator Agent ([generator_model]) produces images based on [user_prompt] with [style_parameters]. The Validator Agent ([validator_model]) independently evaluates each candidate against [realism_criteria], [artifact_detection_rules], and [semantic_consistency_standards]. The pipeline iterates until both agents return TRUE validation signals, ensuring only the highest quality, most realistic outputs are delivered.
</context>

<constraints>
- Both agents must independently return TRUE for final acceptance
- Maximum [max_iterations] generation-validation cycles
- Validator must provide specific rejection reasons for each failed attempt
- Generator must incorporate validator feedback in subsequent iterations
- All outputs must meet [minimum_quality_threshold] standards
- Document the consensus decision trail for auditability
</constraints>

<format>
Return a structured validation report containing:
1. Final approved image (base64 or reference)
2. Consensus confirmation: "GENERATOR: TRUE | VALIDATOR: TRUE"
3. Iteration history with validator feedback
4. Final quality metrics scorecard
5. Generation parameters used
</format>

<tone>
Precise, systematic, quality-obsessed, and collaboratively rigorous
</tone>

<placeholders>
- [generator_model]: e.g., "Midjourney v6", "DALL-E 3", "Stable Diffusion XL"
- [validator_model]: e.g., "CLIP-based aesthetic scorer", "Custom realism classifier", "GPT-4V vision analysis"
- [user_prompt]: The creative prompt describing the desired image
- [style_parameters]: Style, lighting, composition, technical specs
- [realism_criteria]: Texture fidelity, lighting physics, anatomical correctness, etc.
- [artifact_detection_rules]: Watermarks, distortions, hallucinations, inconsistencies
- [semantic_consistency_standards]: Prompt adherence, logical coherence, spatial reasoning
- [max_iterations]: Integer (recommended: 5-10)
- [minimum_quality_threshold]: Score threshold (e.g., 0.85/1.0)
</placeholders>

<final_instruction>
Initialize the dual-agent pipeline now. Begin with Generator Agent producing the first candidate based on [user_prompt], then invoke Validator Agent for assessment. Continue iterative refinement until unanimous TRUE consensus is achieved or [max_iterations] is reached. Output the final validation report.
</final_instruction>