| Generator | Prompt | Labels | Medical | Concept | Polish | Total |
|---|---|---|---|---|---|---|
| Ghost (Human Baseline) | 5 | 5 | 5 | 5 | 5 | 25 |
| AIMAGE | 4 | 5 | 4 | 4 | 4 | 21 |
| Sora | 4 | 5 | 3 | 4 | 4 | 20 |
| Midjourney | 2 | 5 | 1 | 2 | 2 | 12 |
| Flux 2 | 1 | 5 | 1 | 2 | 2 | 11 |
| Nano Banana | 1 | 5 | 1 | 1 | 3 | 11 |
| Grok | 1 | 5 | 1 | 1 | 1 | 9 |
Hits every demanded element cleanly. Correct bony landmarking and cancellous cut, appropriate muscle context, two anchors, two pads, and a clear X-pattern tape construct compressing the footprint. The background treatment matches the "clean studio with wave shapes" spec. This is what the prompt is trying to reproduce, and it's the only image in the set that's a verified medical communication asset rather than a best-effort guess.
Closest to Ghost's composition. Clean white studio look, believable anatomy, background waves present, and the hardware is mostly in the right "family" (pads, tapes, anchor-like elements). Lost points because the tape routing and compression logic doesn't read as a crisp "two wide tapes forming an X over two pads." It's close, but the construct looks slightly improvised, which is exactly where medical device and procedure visuals can become non-defensible.
Strong studio render quality and generally correct "rotator cuff repair vibe." Anchor-like hardware and sutures look plausible, and the tendon-on-bone concept is communicated clearly. Lost points because instead of two wide tapes crossing as an X over two pads, it trends toward multiple passes/loops. It also doesn't clearly deliver the abstract wave background requirement in the same way Ghost does.
High visual punch and material detail. The repair construct becomes fantasy hardware. The blue "mass" and wiring look more like stylized textile or device debris than two tapes + two pads + two anchors. Background is not the clean white studio wave setup, and the anatomy-read is not anchored to a believable surgical illustration.
Clean background, decent rendering polish. It doesn't depict a rotator cuff repair construct. The scene reads like a shoulder anatomy render with an unrelated strap/cap concept. The required anchors, pads, and X-pattern tape compression are not present, so it can't score on adherence or medical accuracy.
Looks dramatic and "medical cinematic." It ignores the studio requirement and substitutes a stylized environment. It also doesn't deliver the specific repair construct (pads, anchors, correct tape layout). This is concept art, not a reproducible surgical visual.
Wrong scene language and framing, wrong background, and the construct is not the requested repair. It doesn't follow the spec closely enough to earn points beyond "no labels."