Hyperrealism ignores eye-level camera constraint, persistent low-angle output, NanoBanana Pro not viable due to facial geometry drift

Hello Support,

I am reporting a persistent issue with camera perspective control in the Hyperrealism model.

Problem:
Hyperrealism repeatedly ignores the “eye-level camera” constraint and generates a low-angle perspective, often closer to chest or belly-button level. This creates a hero-shot look and distorts proportions, which breaks the intended realism for a medium shot.

Why NanoBanana Pro is not an acceptable workaround:
Support suggested toggling NanoBanana Pro in another thread related to a different issue. I tested it, but NanoBanana Pro produces noticeably worse facial geometry for my subject, with uncanny drift and reduced identity consistency.

Hyperrealism is the only mode that reliably preserves facial similarity for my use case, so the issue needs to be addressed in Hyperrealism.

What I tried (among other things):

• Explicitly stated “camera at eye level to the model” multiple times in plain text.

• Repeated the same requirement inside the JSON block using "camera_position": "at eye level".

• Tested multiple focal lengths, including standard portrait ranges (65–85mm) and wider lenses down to 24mm.

• Enabled and disabled close-up mode to check whether framing affected perceived camera height.

• Varied framing (medium shot, waist-to-head, tighter crops) while keeping perspective constraints constant.

• Adjusted subject distance to camera to reduce perspective distortion.

• Avoided wide-angle distortion intentionally and also tested it to rule out lens-related bias.

• Generated multiple batches to account for variance.

Despite these changes, Hyperrealism still frequently outputs a low-angle camera position rather than a neutral eye-level perspective.

Expected behavior:
When “camera at eye level” is requested (in text and JSON), the model should align the camera height with the model’s eyes so the perspective is neutral, not low-angle.

Actual behavior:
Hyperrealism often outputs a camera position below eye level (chest/belly height), even when explicitly specified.

Prompt used (full prompt for reproducibility):
Soft, diffused overcast daylight with gentle, even shadows and low contrast. Medium shot of the model. Model is visible from waist to top of head. Camera is at eye level to the model. Model is wearing a plain muted midnight blue crewneck t-shirt, no logos or graphics. The t-shirt fits naturally, not tight and not oversized. Model is wearing medium grey straight-cut jeans with a clean, minimal look, no distressing or visible branding. Model is looking slightly off camera and gesticulating with his hand as if model is reacting to someone next to the camera. Model’s expression is calm and attentive, with slight cognitive engagement rather than a smile.

{
"subject": {
"distance_to_camera_m": 1.2
},
"pose": {
"body_angle": {
"azimuth_degrees": 30,
"elevation_degrees": 0
},
"eyes": {
"color": "hazel",
"shape": "almond",
"expression": "calm",
"gaze_direction": {
"relative_to_camera": "off_center_left",
"left_of_camera_degrees": 15
}
}
},
"expression": {
"state": "calm",
"engagement_level": "attentive",
"micro_movement": "relaxed face with mild engagement",
"smile": false,
"gaze_direction": {
"relative_to_camera": "off_center_left",
"degrees_left_of_camera": 15
}
},
"photography": {
"camera_setup": {
"lens": "65mm prime",
"aperture": "f/2.8",
"shutter_speed": "1/200s",
"iso": "400",
"sensor": "full frame",
"camera_position": "at eye level",
"focus_point": "eyes"
}
}
}

Any guidance on possible workarounds or future improvements regarding this behavior would be appreciated.