[论文解读] SCHEMA for Gemini 3 Pro Image: A Structured Methodology for Controlled AI Image Generation on Google's Native Multimodal Model

Cazzaniga, Luca|arXiv (Cornell University)|Feb 21, 2026

Data Visualization and Analytics被引用 0

一句话总结

SCHEMA 提供一个三层结构化提示框架（BASE、MEDIO、AVANZATO），适用于 Gemini 3 Pro Image，包含故障路由决策树以及在多个领域经过从业者验证的性能证据。

ABSTRACT

This paper presents SCHEMA (Structured Components for Harmonized Engineered Modular Architecture), a structured prompt engineering methodology specifically developed for Google Gemini 3 Pro Image. Unlike generic prompt guidelines or model-agnostic tips, SCHEMA is an engineered framework built on systematic professional practice encompassing 850 verified API predictions within an estimated corpus of approximately 4,800 generated images, spanning six professional domains: real estate photography, commercial product photography, editorial content, storyboards, commercial campaigns, and information design. The methodology introduces a three-tier progressive system (BASE, MEDIO, AVANZATO) that scales practitioner control from exploratory (approximately 5%) to directive (approximately 95%), a modular label architecture with 7 core and 5 optional structured components, a decision tree with explicit routing rules to alternative tools, and systematically documented model limitations with corresponding workarounds. Key findings include an observed 91% Mandatory compliance rate and 94% Prohibitions compliance rate across 621 structured prompts, a comparative batch consistency test demonstrating substantially higher inter-generation coherence for structured prompts, independent practitioner validation (n=40), and a dedicated Information Design validation demonstrating >95% first-generation compliance for spatial and typographical control across approximately 300 publicly verifiable infographics. Previously published on Zenodo (doi:10.5281/zenodo.18721380).

研究动机与目标

• Address the gap between generic prompt guidelines and production-grade needs for Gemini 3 Pro Image.
• Develop a structured, modular prompt framework with progressive control levels.
• Document model limitations and provide explicit failure routing to alternative tools.
• Empirically validate prompt effectiveness across multiple professional domains.
• Demonstrate information design capabilities with high spatial and typographical control.

提出的方法

• Three-tier progressive structure (BASE, MEDIO, AVANZATO) mapping to control from ~5% to ~95%.
• Seven core labels plus five optional labels for modular prompts.
• Explicit Mandatory and Prohibitions constraints defined with objectively verifiable specifications (e.g., HEX colors, Kelvin temperatures).
• Integrated decision tree with seven questions and three routing exits to alternative tools when Gemini is not suitable.
• Cross-functional features (Thinking Mode, Reference Images, Grounding) available at AVANZATO level to enhance complex scenes.
• Empirical data collection from 850 verified API predictions and ~4,800 generated images across real production contexts.

实验结果

研究问题

• RQ1 Can a model-specific, practitioner-validated structured prompt framework improve consistency and compliance in Gemini 3 Pro Image outputs?
• RQ2 Does a three-tier progressive control approach yield measurable gains in batch coherence and deliverable reliability?
• RQ3 How do constraint-based (Mandatory/Prohibitions) prompts compare to purely descriptive prompts in professional image generation?
• RQ4 What are the model limitations of Gemini 3 Pro Image, and can explicit failure routing mitigate impact?
• RQ5 Is Information Design (spatial layout and typography) reliably achievable with structured prompts in this framework?

主要发现

• Mandatory compliance is 91% and Prohibitions compliance is 94% across domains, with Prohibitions generally outperforming Mandatory.
• SCHEMA AVANZATO prompts yield substantially higher inter-generation coherence than equivalent unstructured prompts in batch tests.
• Independent practitioner validation (n=40) confirms progressive control scaling from BASE to AVANZATO.
• Information Design validation shows >95% first-generation compliance for spatial and typographical control across ~300 publicly verifiable infographics.
• A comparative analysis finds no existing framework that combines model-specific practitioner validation, progressive control, constraint-based specification, and integrated failure routing at production scale.