Studio Product Photography: Evaluating AI Model Performance Across Camera Angles

Research Series: AI Model Comparison for Jewelry Photography

• ← Part 1: Baseline Capability Test
• ← Part 2: Head-to-Head Model Comparison
• Part 3: Studio Shots Comparison (this article)

Abstract

Studio product photography presents distinct challenges from on-hand jewelry shots: precise camera angle control replaces hand realism as the primary technical hurdle. We evaluate 4 frontier models—Nano Banana Pro, Nano Banana (Google), FLUX.2 Pro, FLUX.2 Max (Black Forest Labs)—across 60 images spanning 5 camera angles and 3 ring complexity levels. Our findings reveal that Generate workflow achieves 77% publishable rate versus 53% for Replace, with Generate also producing superior ring accuracy (100% vs 83% good rate). Nano Banana Pro emerges as the clear leader with 93% publishable rate, but Nano Banana exhibits a critical failure mode on Circle View angles (0% accuracy). These results suggest different model selection criteria for studio versus on-hand photography.

1. Introduction

1.1 The Problem

Jewelry e-commerce requires multiple shot types beyond the on-hand images evaluated in Parts 1 and 2 of this research series. Studio shots—product-only images on neutral backgrounds—serve distinct purposes:

• Hero shots: Primary product display, typically 3/4 elevated angle
• Flat lays: Top-down angles for styled compositions
• Detail shots: Specific angles showing ring opening, band profile, or setting details

These shots demand precise camera angle control, a requirement largely absent from on-hand photography where hand positioning provides most of the compositional direction.

1.2 Research Question

We address two questions:

1. Angle control: Can AI models reliably produce studio shots at specific camera angles?
2. Workflow comparison: Does the Generate vs Replace performance pattern from Part 2 hold for studio shots?

1.3 Key Findings Preview

• Generate workflow dominates studio shots (77% vs 53% publishable)
• Ring accuracy is perfect in Generate (100% good) vs 83% in Replace
• Nano Banana Pro achieves 93% publishable rate (highest across all conditions)
• Nano Banana completely fails Circle View angle (0% accuracy across all attempts)

2. Methodology

2.1 Models Evaluated

Based on Part 2 results, we selected the top performers from each workflow:

Generate Workflow:

Replace Workflow:

2.2 Test Design

60 images total:

• 4 models
• 3 rings (simple, medium, complex)
• 5 camera angles
• 1 generation per condition

Camera Angles:

2.3 Evaluation Criteria

Each image was rated blind (evaluator could not see model or workflow) on four dimensions:

2.4 Prompts

Generate prompts used synonym-stacked JSON format optimized in prior angle control research:

{
  "shot_type": ["three quarter view", "3/4 view", "hero shot"],
  "subject": ["this ring", "the ring from reference"],
  "placement": ["standing upright", "ring standing"],
  "camera_angle": ["45 degrees", "elevated front"],
  "style": ["product photography", "e-commerce"]
}

Replace prompts instructed models to swap rings while preserving template composition.

3. Results

3.1 Model Rankings

Nano Banana Pro achieves the highest publishable rate and quality score, though FLUX.2 Max edges ahead on raw angle accuracy.

3.2 Workflow Comparison

Generate workflow wins on every metric except raw angle accuracy. The 17-point ring accuracy advantage is particularly notable—Replace workflow introduces variation, with 17% of outputs rated “similar” (noticeably different from reference).

3.3 Angle Performance

Hero 3/4 View is reliably produced by all models. Circle View shows the highest failure rate at 25%.

3.4 The Nano Banana Circle View Failure

Nano Banana fails Circle View completely—all three attempts scored 1 (failed). The model does not interpret “see into ring hole” correctly. This represents a categorical failure mode absent from the Pro version.

3.5 Ring Accuracy

By Model:

By Workflow:

Generate workflow preserves ring design perfectly. Replace introduces variation—17% of rings came out merely “similar.”

3.6 Cost Efficiency

Nano Banana offers best value at $0.065 per publishable image—but the Circle View limitation may disqualify it for some catalogs.

4. Analysis

4.1 Why Generate Wins for Studio Shots

In Part 2 (on-hand shots), Replace workflow showed advantages for ring accuracy because BFL models excel at image editing. For studio shots, this advantage disappears:

1. No hand to preserve: Replace workflow’s strength is maintaining hand consistency. Studio shots have no hand.
2. Template dependency: Replace requires pre-generated templates. Any template issues propagate.
3. Ring placement complexity: Replacing a ring in a specific pose/angle is harder than generating around a ring.

Generate builds the composition around the reference ring. Replace tries to edit a ring into an existing composition—introducing more opportunities for ring accuracy errors.

4.2 The Quality-Accuracy Trade-off

Nano Banana Pro offers the optimal balance. FLUX.2 Max achieves marginally better angle accuracy but substantially lower quality and publishable rate.

4.3 Failure Mode Analysis

Only one image had a tagged issue (FLUX.2 Max: “AI-obvious” on complex flat lay). This is remarkably clean compared to Part 2’s on-hand shots.

Interpretation: Studio shots eliminate the hand-related failure modes that dominated on-hand photography. No hands = no hand problems.

5. Discussion

5.1 Part 2 vs Part 3 Comparison

Studio shots are easier for AI than on-hand shots. The absence of hands eliminates the primary source of failures and AI artifacts.

5.2 Implications for Model Selection

For studio product photography:

• Use Generate workflow with Nano Banana Pro
• Nano Banana is viable for budget applications but avoid Circle View
• Replace workflow offers no advantages over Generate for this use case

For on-hand photography (Part 2 findings):

• Generate: Google models (NBPro)
• Replace: BFL models (FLUX.2 Max, FLUX.2 Pro)

The optimal model depends on shot type, not just workflow preference.

5.3 The Nano Banana Limitation

Nano Banana’s Circle View failure represents a categorical model limitation. At $0.039/image (3.8× cheaper than Nano Banana Pro), it offers compelling value—but only for angle types it can reliably produce.

Recommendation: Route Circle View requests to Nano Banana Pro; use Nano Banana for other angles where budget is constrained.

6. Limitations

• Single evaluator (blind rating reduces bias)
• One generation per condition (no variance testing)
• Ring styles may not generalize to all jewelry types
• December 2025 model versions
• Templates generated with Nano Banana Pro (may favor Generate comparison)

7. Conclusion

For studio product photography, Generate workflow with Nano Banana Pro is the clear recommendation:

1. 93% publishable rate (highest across all conditions)
2. 100% ring accuracy (perfect preservation of reference)
3. 4.73 quality average (highest visual appeal)
4. No angle failures (reliable across all 5 angles)
5. $0.161 per publishable (reasonable premium for top quality)

The same model that won Parts 1 and 2 continues to dominate. However, the workflow dynamics differ: while on-hand shots showed genuine trade-offs between Generate and Replace, studio shots favor Generate decisively.

For budget-constrained applications, Nano Banana achieves $0.065 per publishable image but must be excluded from Circle View shots.

Research conducted December 2025. 60 images evaluated blind across 4 models and 5 camera angles.

studio formel Research — Advancing AI for commercial content generation.

Related Articles

• The Complete Guide to Jewelry Photography — All shot types including studio shots
• Prompt Engineering: Angle Control — How to control camera angles in prompts
• Which AI Model Works Best for Jewelry Photography? — Practical model recommendations

About studio formel

studio formel is an AI-powered creative platform built specifically for jewelry brands. We combine systematic research on AI generation with a flexible asset management system, helping jewelry sellers create professional images, videos, and ads at scale.

Learn more about our platform →