Summary for GPT Image 1.5
GPT Image 1.5 emerges as a solid mid-tier performer, securing an overall Elo rating of 1095. While it operates behind current state-of-the-art leaders like GPT Image 2.5 Sunburst (1348 Elo), it offers notable improvements and creative capabilities well above older baseline models like DALL-E 3 (780 Elo).
Key Findings:
- Surprising Strengths: It performs exceptionally well in complex and surreal tasks, achieving its absolute highest Elo rating (1286) in the Ultra Hard category.
- Aesthetic Bias: The model leans heavily toward over-processed, "plastic," or heavily 3D-rendered aesthetics. This dramatically hurts its performance in tasks requiring flat, minimalistic, or strictly photographic outputs.
- Aggressive Safety Filters: A major trend is its high refusal rate (11 out of 100 prompts). It completely collapsed in the anime and Ghibli categories due to over-sensitive safety triggers.
- Quick Takeaway: Leverage GPT Image 1.5 for highly imaginative, textured, and surreal concepts. Avoid using it for clean graphic design, flat vectors, or any prompt that might trigger sensitive safety filters.
General Analysis & Useful Insights
Strengths & Differentiators
- Imaginative Execution: GPT Image 1.5 thrives on visual density and creative combination. It frequently scores 8s and 9s in detail execution when tasked with rich textures, such as the Floating steampunk castle (Score 9).
- Strong Subject Focus: The model consistently adheres to core subjects and action requests. It successfully renders nuanced concepts like the American Sign Language gesture with excellent anatomical clarity (Score 9), resulting in the ASL 'thank you' image.
- High Contrast Lighting: The model excels at dramatic lighting and atmospheric depth, making scenes feel cinematic and impactful.
Weaknesses & Failure Modes
- The "Over-Processed" Look: A recurring failure mode is aggressive surface sharpening and glossy highlights. In the Photorealistic People & Portraits category, skin often looks beauty-smoothed or hyperreal, severely limiting its photographic credibility. For instance, the Old fisherman portrait suffered significantly due to exaggerated skin treatments.
- Graphic Design Struggles: The model fundamentally struggles with minimalism. When asked for flat vectors or clean designs, it forcibly injects 3D lighting, glows, and vignettes. This is painfully obvious in the Minimalist coffee logo, which received a dismal score of 3.
- Spatial Clutter in Complex Scenes: When managing multiple overlapping elements, the model often merges fine details into visual noise, as seen in the slightly muddled background of the Bustling market scene.
High Refusal Rate
The safety filter is uniquely aggressive. 11% of all prompts were rejected, primarily in stylized or anime categories. For example, Yoga practitioner pose and Magical bathhouse were completely blocked, limiting the model's reliability in automated content pipelines.
