Commercial text-to-image models produce accurate images for broad nuclear concepts but fail on specialized engineering prompts¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? GPT-Image-2 and Gemini-3.1-Flash-Image were significantly better than Midjourney on nuclear concepts across nearly all 300 images, yet all three commercial models failed to generate accurate images as prompts became specific. → Mohammed I. Radaideh 2026

Evidence¶

Mohammed I. Radaideh 2026¶

Mohammed I. Radaideh, Jeremy Moon, Andre Gala-Garza, Emma Son, Yug Shah, Majdi I. Radaideh. (2026). NuclearDiffusion: Text-to-Image Foundation Models for Learning Nuclear Energy Concepts. Preprint submitted to Peer Review. https://arxiv.org/abs/2608.04030 doi:10.48550/arxiv.2608.04030

q2 · i? · design · r2

Section 6 comparison of GPT-Image-2, Gemini-3.1-Flash-Image, and Midjourney on 300 testing prompts, with qualitative inspection (e.g., the TVS-2M VVER-1000 fuel assembly prompt in Figure 9) and CMMD/KID in Table C.1. The "significantly better" GPT/Gemini versus Midjourney contrast is reported without an effect size.

"However, as the prompts became specific, all three failed to generate accurate images, suggesting the necessity for fine-tuning."

Discussion¶

Related Claims¶

• Generating multiple images per prompt and re-prompting with word variations did not yield accurate nuclear images for failed concepts — related
• LoRA fine-tuning of Flux.1 failed to learn nuclear concepts, producing wrong-concept images on all 300 test prompts — related
• Human qualitative assessment shows SDXL significantly outperforms SD-v3.5-Medium after fine-tuning, failing on 84 versus 250 of 300 prompts — related
• The CLIP text-image alignment score failed to capture fine-tuning improvements, contradicting KID, CMMD, and human judgment — related
• Fine-tuning on the 1,000-image nuclear dataset improves image fidelity of SD models, with KID and CMMD decreasing over training steps — related
• One student group compared outputs across three LLMs and judged Gemini more accurate overall, though it still failed fine-grained distinctions — related