cs.CVFeb 21, 2026

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

Authors: Luca Cazzaniga

Organizations: Independent Researcher — AI-Assisted Visual Production

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

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).

Figures & tables

FeatureBAS (Lee & Park, 2025)SCHEMA (Cazzaniga, 2026)
Target AI ModelAgnostic (generic visual AI)Specific (exclusively Gemini 3 Pro Image)
Application DomainBiophilic Architecture (single domain)Real Estate, Product, Editorial, Storyboard, Advertising, Information Design (multi-domain)
Prompt Architecture5 fixed linear components7 core labels + 5 optional labels, highly modular
Constraint ManagementGeneric Negative promptMandatory and Prohibitions based on objectively verifiable metrics
Failure RoutingNot contemplatedExplicit integrated decision tree
Empirical BasisExpert evaluation (qualitative)621 prompts, 850 verified API predictions, n=40n=40 workshop validation, ∼\sim 300 publicly verifiable infographics
Table 1: Comparative synopsis: BAS Framework vs. SCHEMA
ParameterBASE (Discovery)MEDIO (Direction)AVANZATO (Deliverable)
Practitioner Control∼\sim 5%∼\sim 85%95–98%
AI Creativity∼\sim 95%∼\sim 15%≤\leq 5%
Time Investment<< 1 minute∼\sim 5 minutes>> 15 minutes
Prompt StructureFree natural language7 structured labels7 core + 5 optional labels
PurposeExploration, bias identificationProfessional draftsFinal deliverables, batch consistency
Table 2: SCHEMA three-tier progressive structure. Control and creativity percentages are design targets qualitatively confirmed by independent practitioner observation ( n=40n=40 , Section 5.2 ) and indirectly supported by batch consistency data (Section 5.1 ).
LabelFunctionKey Specification
SubjectUnambiguous definition of the main scene elementPrecise materials, dimensions, exact colors, state
StyleGlobal aesthetic reference anchoringPhotography type, quality level, brand or editorial reference
LightingTotal illumination environment controlPhotographic setup, angle of incidence, Kelvin temperature (numeric)
BackgroundSurrounding environmental contextSpatial setting, surface materials, depth of field
CompositionVirtual camera direction and perspectiveShot type, shooting angle, focal point definition
MandatoryEnsuring presence of critical elements (3–10 items)Exclusively objectively verifiable technical or visual items
Table 3: SCHEMA seven-label core architecture
DomainPromptsExec.ImagesMand.Prohib.Failure
 (n)(est.)(est.)Compl.Compl.Routing
Real Estate62∼\sim 362∼\sim 42491%95%n≈2n\approx 2
Product217∼\sim 1,369∼\sim 1,77491%95%n≈7n\approx 7
Editorial149∼\sim 957∼\sim 1,24592%95%n≈4n\approx 4
Storyboard41∼\sim 258∼\sim 30289%91%n≈2n\approx 2
Campaign77∼\sim 541∼\sim 77090%92%n≈4n\approx 4
Table 4: Aggregated corpus data by professional domain. Compliance rates are practitioner-assessed; execution and image counts are retrospective estimates except for the 850 verified Replicate API predictions and the ∼\sim 300 publicly verifiable Information Design images.
DomainCondition ACondition BGap
 (Narrative)(SCHEMA AVANZATO) 
Real Estate4–5/108–9/10+4
Product3–5/108–9/10+4
Editorial4–6/108–9/10+3.5
Storyboard3–4/107–8/10+4
Campaign3–5/108–9/10+4
Table 5: Comparative Batch Consistency Test results. Consistency Score = mean number of substantially identical images per 10-image batch, averaged across trials and raters.
IterationObserved EffectPractical Notes
Generation 1 (original)Optimal qualityReference baseline. Always save before iterating.
Generation 2 (1st edit)Initial visible degradationNoisier pixels, micro-artifacts visible at magnification. Acceptable for commercial use.
Generation 3 (2nd edit)Moderate degradationSpurious lines, edge sharpness loss, slight color shift. Partially mitigable by reloading at 4K.
Generation 4+ (3rd edit+)Severe degradationEvident defects, smudging, critical detail loss. Not production-ready without post-processing.
Table 6: Iterative Generative Drift: observed progression
LabelContentMethodological Notes
StyleProfessional architectural interior photography, high-end real estate magazine quality, photorealisticDefines visual register and expected quality standard
CompositionWide-angle frontal view at eye level, symmetrical framing, 16–24mm equivalent focal lengthExplicit focal length and viewpoint reduce compositional ambiguity
SubjectModern living room with light oak flooring, white walls, large sofa in warm grey fabric, coffee tableSpecific materials and colors, not generic descriptors
LightingNatural soft daylight from large window, warm 3000K recessed ceiling lights, 70:30 natural to artificial ratioNumeric color temperature + light source ratio
MandatoryVerticals perfectly straight, realistic materials, no perspective distortion, professional photography standardsPositive constraints—hardest to enforce (91% compliance)
ProhibitionsNo converging verticals, no oversaturated colors, no fake HDR, no unrealistic furnitureNegative filters—highest compliance (95%)
Table 7: MEDIO prompt structure: Real Estate Interior
LabelContentMethodological Notes
StyleHigh-end commercial advertising photography, luxury brand campaign aesthetic, hyperreal cinematic polish, aspirational qualityMultiple style qualifiers increase visual register precision
ReferenceImage 1: Product identity source — maintain 100% accuracy in shape, proportions, label design, typography, exact color values, branding elements, material textureDedicated Reference section with explicit fidelity instruction—key for Product compliance
Composition3 ×\times 3 grid layout: Cell 1 hero still life, Cell 2 extreme macro detail, Cell 3 dynamic liquid interaction, Cell 4 minimal sculptural arrangement, Cell 5 floating elements, Cell 6 sensory close-up, Cell 7 color-driven scene, Cell 8 ingredient abstraction, Cell 9 surreal fusionEach cell named individually—reduces compositional drift between frames
ThinkingEnable: Multi-stage refinement for complex 9-cell composition. Priority: 95% accuracy on product replication, 5% creative concept. Complexity: HighAVANZATO-exclusive section activating multi-step reasoning
MandatoryProduct 100% accurate in every cell: exact shape, proportions, label typography, color matching. Ultra-sharp focus. Premium editorial quality consistent throughout gridMandatory repeated contextually for each grid cell
ProhibitionsNo product distortion or alterations. No inconsistent style between cells. No generic commercial photography. No color treatments diverging from reference. No amateur post-production quality5 specific prohibitions vs. generic ones—specificity increases compliance
Table 8: AVANZATO prompt structure: Product Campaign with Reference
More in this topic

Prompt Optimization for T2I Generation

Prompt optimization for text-to-image generation improves how natural-language instructions control image content, style, composition, and quality.

See all 31 papers

Explore similar work

CardsList
Aug 15, 2026

Benchmarking Frontier Text-to-Image Models on Image-Description Prompts

Text-to-image models are typically reported on average-case prompts, which understates the gap between systems on compositionally demanding requests involving precise object counts, multi-object attribute binding, legible embedded text, and explicit spatial constraints. We evaluate four production text-to-image systems: Hunyuan 3.0, Gemini 3 Pro Image ("Nano Banana Pro"), Black Forest Labs FLUX.2, and Ideogram 3.0. The evaluation uses the 48 hardest prompts drawn from the DataSeeds.AI Sample Dataset (DSD), selected through an automated complexity-scoring pass over the full corpus. Every generated image is graded using an independent-judge rubric. GPT-5.4-Pro authors an atomic, weighted, mutually exclusive and collectively exhaustive (MECE) evaluation rubric, while Gemini 3.1 Pro Preview independently determines whether each criterion is satisfied. Gemini 3 Pro Image ranks first with a score of 84.8/100, narrowly ahead of FLUX.2 at 82.3/100. Ideogram 3.0 and Hunyuan 3.0 score 65.7/100 and 63.3/100, respectively. Failure analysis shows that the leading systems primarily lose points through object miscounting and geometric artifacts, whereas the trailing systems more frequently produce garbled text. Ideogram 3.0 also frequently omits requested elements. Full per-sample rubrics, scores, and failure annotations are available from the authors upon request.

Sajjad Abdoli, Ghassan Al-Sumaidaee, Ahmed Rashad

Perle

Text Rendering in T2I GenerationT2I GenerationT2I Generation EvaluationCompositional T2I Generation
AskPDF
Oct 16, 2026

Agentic Image Generation in the Wild

Agentic image generation pairs an image generator with a search agent that retrieves references from the web, extending generation beyond the generator's training data. Real-world requests, however, often concern entities that appeared only days ago. We call this setting Image Generation in the Wild. In this setting, the web offers limited visual evidence, search results are crowded with easily confused entities, and the target is rarely shown on its own. To study this setting, we build an automated pipeline that synthesizes prompts from recent headlines and keeps only those whose real search results are highly confusing, yielding requests that are hard to solve yet natural to ask. With it, we construct Savanna, a benchmark of 500 prompts across 16 subcategories. On Savanna, even harness frameworks built on the frontier models retrieve a correct reference for only about half of the prompts. We then propose Locate-Corroborate-Isolate, a paradigm in which the agent first pins down the target's identity through text search, then verifies candidate images against multiple sources, and finally crops a clean reference for the generator. Trained on 15K trajectories that follow this paradigm, Ranger-8B achieves a search score of 67.2, outperforming harnesses equipped with frontier models. It also achieves the best generation quality with every generator we test, and surpasses prior agentic methods on other image generation benchmarks. We release the pipeline, benchmark, training data and model at https://github.com/RUCAIBox/Agentic-ImageGen-Wild.

Yifan Li, Bowen Qin, Kun Zhou +8

Renmin University of China · Tencent · Peking University

Agentic Image Generation
AskPDF
May 8, 2026

SCOPE: Structured Decomposition and Conditional Skill Orchestration for Complex Image Generation

While text-to-image models have made strong progress in visual fidelity, faithfully realizing complex visual intents remains challenging because many requirements must be tracked across grounding, generation, and verification. We refer to these requirements as semantic commitments and formalize their lifecycle discontinuity as the Conceptual Rift, where commitments may be locally resolved or checked but fail to remain identifiable as the same operational units throughout the generation lifecycle. To address this, we propose SCOPE, a specification-guided skill orchestration framework that maintains semantic commitments in an evolving structured specification and conditionally invokes retrieval, reasoning, and repair skills around unresolved or violated commitments. To evaluate commitment-level intent realization, we introduce Gen-Arena, a human-annotated benchmark with entity- and constraint-level specifications, together with Entity-Gated Intent Pass Rate (EGIP), a strict entity-first pass criterion. SCOPE substantially outperforms all evaluated baselines on Gen-Arena, achieving 0.60 EGIP, and further achieves strong results on WISE-V (0.907) and MindBench (0.61), demonstrating the effectiveness of persistent commitment tracking for complex image generation.

Tianfei Ren, Zhipeng Yan, Yiming Zhao +13

MoE Key Laboratory of Brain-inspired Intelligent Perception and Cognition, University of Science and Technology of China · The Hong Kong Polytechnic University · Nanyang Technological University

Conditional Image GenerationCompositional T2I GenerationT2I Generation Evaluation
AskPDF