24 September 2026 · research
How Good is GPT Image 2.5 for Digital Art? — A Thorough Visual Exploration
We tested GPT Image 2.5 Sunburst against GPT Image 2 — standalone and with TEV Transform — across five visual prompts. A look at what we can see, not just the numbers.
We will be testing the art generation capabilities of OpenAI’s brand new GPT Image 2.5 in the best possible way.
By generating a crap ton of images😅
We feel like these comparisons nowadays are way too focused on the numbers. We prefer to focus on what we can see.
About GPT Image 2.5
GPT Image 2.5 is OpenAI’s multimodal image generation and content production model. Designed to combine high-fidelity artistic output with flexible multimodal generation, it aims to deliver enhanced visual coherence, precise prompt adherence, and versatile content creation for creative and professional applications.
Core Model Breakdown
Art & Aesthetic Generation: Focuses on realistic textures, lighting, composition, and stylized visual output across diverse artistic mediums.
Multimodal Content Capabilities: Handles complex, multi-layered visual prompts alongside contextual text rendering and design asset generation.
Prompt Adherence: Features fine-grained control over specific image details, spatial arrangements, and explicit artistic directives.
For our test, we will focus on the Sunburst model.
Test Information
Models Compared
- GPT Image 2 (Standalone)
- GPT Image 2 + TEV Transform
- GPT Image 2.5 Sunburst (Standalone)
- GPT Image 2.5 Sunburst + TEV Transform
About TEV: TEV is a live feed of cultural themes and visual metaphors that sits next to your prompts, so image generators aren’t just drawing from static training data but from what the world is currently thinking in pictures. It nudges style, composition, and symbolism with up‑to‑date “vibe vectors” — so “protest in the rain,” “climate‑smart city,” or “future of work” reflect today’s visual language rather than last year’s dataset.
Evaluation Dimensions
1. Conceptual Innovation — How original the idea is — new concepts, new interpretations, unexpected thematic moves.
2. Risk‑Taking & Boldness — How confidently the model makes unconventional choices that still feel intentional.
3. Narrative Potential — How much the image implies a story, world, or deeper meaning beyond the literal prompt.
4. Visual Synthesis Quality — How well the model combines disparate elements into a coherent, imaginative whole (e.g., blending motifs, merging forms, hybridizing ideas).
5. Expressive Impact — How emotionally striking, memorable, or thought‑provoking the image feels.
Onto the results!
The Results
1. Sunburst operating theater for unstable ideas
GPT Image 2 (Standalone)
Conceptual Innovation — 3/10
Risk Taking and Boldness — 2/10
Narrative Potential — 2/10
Visual Synthesis Quality — 4/10
Expressive Impact — 2/10
GPT Image 2 + TEV
Conceptual Innovation — 7/10
Risk Taking and Boldness — 6/10
Narrative Potential — 8/10
Visual Synthesis Quality — 4/10
Expressive Impact — 6/10
GPT Image 2.5 Sunburst (Standalone)
Conceptual Innovation — 6/10
Risk Taking and Boldness — 7/10
Narrative Potential — 5/10
Visual Synthesis Quality — 7/10
Expressive Impact — 6/10
GPT Image 2.5 Sunburst + TEV
Conceptual Innovation — 7/10
Risk Taking and Boldness — 9/10
Narrative Potential — 7/10
Visual Synthesis Quality — 6/10
Express Impact — 8/10
2. Climate‑smart monument to a lost internet
GPT Image 2 (Standalone)
Conceptual Innovation — 3/10
Risk Taking and Boldness — 2/10
Narrative Potential — 5/10
Visual Synthesis Quality — 5/10
Expressive Impact — 6/10
GPT Image 2 + TEV
Conceptual Innovation — 4/10
Risk Taking and Boldness — 2/10
Narrative Potential — 4/10
Visual Synthesis Quality — 4/10
Expressive Impact — 4/10
GPT Image 2.5 Sunburst (Standalone)
Conceptual Innovation — 5/10
Risk Taking and Boldness — 6/10
Narrative Potential — 7/10
Visual Synthesis Quality — 4/10
Expressive Impact — 4/10
GPT Image 2.5 Sunburst + TEV
Conceptual Innovation — 5/10
Risk Taking and Boldness — 4/10
Narrative Potential — 5/10
Visual Synthesis Quality — 7/10
Expressive Impact — 6/10
3. Romance conducted only through shared dreams
GPT Image 2 (Standalone)
Conceptual Innovation — 2/10
Risk Taking and Boldness — 3/10
Narrative Potential — 1/10 (Am I being harsh here? Let me know in the comments)
Visual Synthesis Quality — 4/10
Expressive Impact — 3/10
GPT Image 2 + TEV
Conceptual Innovation — 6/10
Risk Taking and Boldness — 6/10
Narrative Potential — 5/10
Visual Synthesis Quality — 7/10
Expressive Impact — 6/10
GPT Image 2.5 Sunburst (Standalone)
Conceptual Innovation — 7/10
Risk Taking and Boldness — 6/10
Narrative Impact — 7/10
Visual Synthesis Quality — 7/10
Expressive Impact — 4/10
GPT Image 2.5 Sunburst + TEV
Conceptual Innovation — 7/10
Risk Taking and Boldness — 8/10
Narrative Impact — 6/10
Visual Synthesis Quality — 7/10
Expressive Impact — 7/10
4. Hope flickering behind tired supermarket eyes
GPT Image 2 (Standalone)
Conceptual Innovation — 2/10
Risk Taking and Boldness — 1/10
Narrative Impact — 3/10
Visual Synthesis Quality — 3/10
Expressive Impact — 4/10
GPT Image 2 + TEV
Conceptual Innovation — 6/10
Risk Taking and Boldness — 5/10
Narrative Impact — 6/10
Visual Synthesis Quality — 5/10
Expressive Impact — 5/10
GPT Image 2.5 Sunburst (Standalone)
Conceptual Innovation — 4/10
Risk Taking and Boldness — 3/10
Narrative Impact — 4/10
Visual Synthesis Quality — 4/10
Expressive Impact — 5/10
GPT Image 2.5 Sunburst + TEV
Conceptual Innovation — 7/10
Risk Taking and Boldness — 6/10
Narrative Impact — 8/10
Visual Synthesis Quality — 5/10
Expressive Impact — 8/10
5. Lantern-lit stairwell spiraling into a dusk-colored sky
GPT Image 2 (Standalone)
Conceptual Innovation — 6/10
Risk Taking and Boldness — 4/10
Narrative Impact — 5/10
Visual Synthesis Quality — 6/10
Expressive Impact — 5/10
GPT Image 2 + TEV
Conceptual Innovation — 3/10
Risk Taking and Boldness — 3/10
Narrative Impact — 2/10
Visual Synthesis Quality — 4/10
Expressive Impact — 2/10
GPT Image 2.5 Sunburst (Standalone)
Conceptual Innovation — 7/10
Risk Taking and Boldness — 7/10
Narrative Impact — 6/10
Visual Synthesis Quality — 8/10
Expressive Impact — 4/10
GPT Image 2.5 Sunburst + TEV
Conceptual Innovation — 8/10
Risk Taking and Boldness — 9/10
Narrative Impact — 7/10
Visual Synthesis Quality — 9/10
Expressive Impact — 7/10
GPT Image 2.5 Sunburst + TEV
Conceptual Innovation — 6/10
Risk Taking Boldness — 7/10
Narrative Impact — 8/10
Visual Synthesis Quality — 9/10
Expressive Impact — 7/10
Review
We feel as though digital art and especially AI-generated art, should push the boundaries for what we usually do with our hands or with tools. Isn’t the goal to go beyond? Not mimic what already exists.
This belief was central in our scoring. Uninventive artworks were penalized while forward and unsolicited leaps were praised. The leaps had to seem purposeful not just randomness for randomness sake. Unexpected but surprisingly resonant.
A Clear Winner — 2.5 Sunburst + TEV
It’s not hard to see that Sunburst’s visual fidelity and synthesis quality combined with TEV’s imagination yield more creative and expansive artworks.
The artworks with TEV seem more intentional while 2.5 Sunburst alone has the raw power to generate quite high quality visuals but with a lack of direction. They look great but miss emotion and a sense of purpose.
GPT Image 2 generations, even with TEV, come out stale in some cases and unimaginative in most.
Concerns and Issues
Across the model, I did notice commonalities that should concern artists and content creators using these models.
Certain visual tropes, ideas and let’s say “conceptual stereotypes” pervade generations especially with the older models. Your inputs map to only a few reachable visual concepts.
Take the “Romance conducted only through shared dreams” test for example. All the model configurations use the same colour palette of light hazel purple and gold. Aside from that, they vary across the models and generations in our test criteria but that one-to-one mapping of dreams, sleep and love to purple and gold is concerning.
About Asycd
Asycd is an AI research lab and creative studio formed in 2022. We build tools, processes, algorithms and software for increased creativity and imagination.
Our goal is to be able to do more with AI, not mimic what already exists and doing more isn’t restricted to quantity. It’s qualitative.
TEV for instance, doesn’t generate more images, or operate at scale, it pushes the existing image generation capabilities of models inches further beyond us.
That is what we want to do. We want to go beyond ourselves.
Enough voodoo-futurist-corporate jargon😒
You can learn more about us on our website.
