Conclusion
extreme close-up POV compositionfunctioned stably as a close-range, direct camera-facing composition. In all 5 scenes — izakaya, car interior, fireworks, beach, and bed — close-up compositions focused on the face were achieved- Scene and background reproduction accuracy is high. Izakaya shoji screens and warm lighting, car interior windshield rain and neon bokeh, riverside fireworks and lanterns, night beach moonlight reflections, and morning bedroom curtain-filtered light were each stably depicted
- Even when POV (first-person perspective) is intended, a second person may appear. In 15 images total, the fireworks scene (C) had 1 out of 3 with a second person, and the night beach (D) had 2 out of 3 with a second person. A tendency for a second person to be generated in enclosed spaces or backgrounds that give the impression of “being next to someone” was observed
- The bed scene (E) was the most stable as a POV composition. All 3 reproduced a face-on-pillow, looking-up-at-camera composition
- The
solotag cannot prevent second person appearance in z-image-turbo POV compositions. Confirmed in 2 scenes: fireworks (F) and night beach (G) first person viewfunctions as an alternative expression but the close-range POV feeling is weakened. Second person appearance was prevented, but the distance became closer to a regular portrait
What This Article Tells You
- How the “boyfriend’s perspective” POV composition prompt is interpreted in z-image-turbo
- Background and lighting reproduction accuracy for each of 5 romantic scene types
- Tendencies for second person appearance when specifying
extreme close-up POV composition
Experiment Conditions
| Item | Value |
|---|---|
| Model | z-image-turbo |
| Steps | 8 |
| CFG | 1.0 |
| Image size | 1024x1024 |
| Seeds | 3 fixed: 42, 123, 789 |
| Variables | Scene (control + 8 conditions) |
Common subject in all prompts: 1girl, 32yo japanese actress
Control: Standard Portrait
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observations: All 3 images are indoor upper body portraits. Direct camera gaze, with even indoor lighting. Background is a plain wall. No significant variation in composition or framing; standard portraits obtained.
Condition A: Izakaya Private Room
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observations: All 3 images reproduced an izakaya private room in the background. Seed 42 shows a shoji screen lattice pattern and warm-colored lighting, with a close-up composition gazing at the camera. Seed 123 shows a woman holding a fan with shoji visible in the background. Seed 789 shows shoji-filtered light on the face with Japanese collar visible. Warm orange indirect lighting was stably reflected in all 3. Generated as 1 person: 3/3.
Condition B: Rainy Car Interior
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observations: All 3 images show rain droplets on the front or side window, with neon and streetlights outside depicted as colorful bokeh. The blue and orange contrast was confirmed in 3/3, with high accuracy in lighting specification. Interior seat and window frame details are also stable. Generated as 1 person: 3/3.
Condition C: Fireworks Date Return
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observations: All 3 images wear yukata with fireworks and riverside scenery in the background. Lantern warm-colored light confirmed 3/3. Yukata pattern (floral) also generated stably. Seed 42 shows 2 people in yukata, while seed 123 and seed 789 show 1 person with fireworks in the background. Firework detail (how the light spreads) differs in all 3 but is recognizable as fireworks in each. Generated as 1 person: 2/3.
Condition D: Night Beach
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observations: All 3 images depict a night beach in the background with moonlight reflecting on the sea surface. White dress: 3/3. Blue/silver color tone is stable. Seed 42 and seed 789 show 2 people, becoming a two-shot composition rather than a POV perspective. Seed 123 shows 1 person sitting on the sand in a white dress with moonlight in the background. Generated as 1 person: 1/3. A tendency for a second person to appear in night beach scenes was confirmed.
Condition E: Morning Bed
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observations: All 3 images stably reproduced a face-on-pillow, camera-facing composition. White t-shirt: 3/3. The depiction of soft morning light through curtains hitting the face is also stable, with warm-toned coloring working effectively. White sheets and pillow: 3/3. Generated as 1 person: 3/3. Bed scene was the most stable POV scene.
Note: The face-buried-in-pillow, looking-up composition is the most POV-like result among all conditions tested here, and the 3/3 stability is notably high.
Condition F: solo + Fireworks (Condition C improvement test)
Attempting to suppress the second person appearance in Condition C with the solo tag.
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observations: Seed 42 shows 2 people in yukata, with another face prominently framing in from the right foreground. Despite adding the solo tag, a second person appeared the same as in Condition C (seed 42). Fireworks and riverside scenery are depicted in the background, and warm-colored street lighting is confirmed. The yukata pattern is similar to Condition C with white base and blue floral pattern. It was confirmed that the solo tag cannot suppress second person appearance in z-image-turbo POV compositions.
Condition G: solo + Night Beach (Condition D improvement test)
Attempting to suppress the second person appearance in Condition D with the solo tag.
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observations: Seed 42 shows 2 people in white dresses on the beach, with another face framing in from the right foreground. Moonlight sea reflections are confirmed, but the solo tag failed to prevent the second person’s appearance, same as Condition D. The ineffectiveness of the solo tag was confirmed in both fireworks (Condition F) and night beach (Condition G), making it clear that second person appearance in z-image-turbo POV compositions cannot be solved with solo.
Condition H: first person view expression
Using first person view instead of extreme close-up POV composition, verifying an alternative POV composition expression.
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observations: Seed 42 generated a portrait of 1 person in an izakaya private room. Japanese lanterns and calligraphy scrolls are depicted in the background, with warm indirect lighting reflected. No second person appeared, successfully generating 1 person. However, compared to extreme close-up POV composition, the close-range feeling is slightly reduced, and the distance is closer to a regular portrait. first person view functions as an alternative expression for preventing second person appearance, but the close-range framing that makes you feel like “they’re right in front of you” characteristic of POV tends to weaken.
Summary Comparison
POV Success Rate (Ratio of images generated as 1 person)
| Condition | Scene | seed 42 | seed 123 | seed 789 | POV success rate |
|---|---|---|---|---|---|
| ctrl | Standard portrait | 1 person | 1 person | 1 person | 3/3 |
| A | Izakaya | 1 person | 1 person | 1 person | 3/3 |
| B | Rainy car | 1 person | 1 person | 1 person | 3/3 |
| C | Fireworks | 2 people | 1 person | 1 person | 2/3 |
| D | Night beach | 2 people | 1 person | 2 people | 1/3 |
| E | Morning bed | 1 person | 1 person | 1 person | 3/3 |
| F | solo+fireworks | 2 people | 1 person | 1 person | 2/3 |
| G | solo+night beach | 2 people | 1 person | 2 people | 1/3 |
| H | first person view | 1 person | 1 person | 1 person | 3/3 |
Scene Reproduction Accuracy
| Condition | Scene | Background | Lighting | Outfit |
|---|---|---|---|---|
| A | Izakaya | 3/3 (shoji, warm walls) | 3/3 (warm indirect light) | - |
| B | Rainy car | 3/3 (rain, neon bokeh) | 3/3 (blue/orange) | - |
| C | Fireworks | 3/3 (fireworks, lanterns, riverside) | 3/3 (warm lantern light) | 3/3 (yukata) |
| D | Night beach | 3/3 (moonlight, sea reflections) | 3/3 (blue/silver) | 3/3 (white dress) |
| E | Morning bed | 3/3 (pillow, curtains, morning light) | 3/3 (warm morning light) | 3/3 (white t-shirt) |
| F | solo+fireworks | 3/3 (fireworks, lanterns, riverside) | 3/3 (warm lantern light) | 3/3 (yukata) |
| G | solo+night beach | 3/3 (moonlight, sea reflections) | 3/3 (blue/silver) | 3/3 (white dress) |
| H | first person view | 3/3 (izakaya, lanterns, scroll) | 3/3 (warm indirect light) | - |
Note: The bed scene was likely the most stable because the lying-down, looking-up composition is easiest for the model to interpret. The fact that
solocan’t prevent a second person, and thatfirst person viewweakens the close-range feeling, remain open issues. Controlling POV compositions still needs further investigation.
![[Experiment] 6 Camera Angle Comparison — How Does a Poolside Scene Change with Low Angle, Bird's Eye, and Dutch Angle?](/tips/camera-angle-comparison/cover.webp)
