Production breakdown / Nova Grande
Less render.
More photograph.
Keep the character. Change how light, materials and the camera describe her.
- My contribution
- Reference selection, scene direction, prompting and visual QA
- Tools used
- Native ChatGPT / OpenAI image generation
- Output
- Three lighting tests, followed by six profile variations
The shift in treatment
Different scenes from the same character project. This shows the direction of the iteration, rather than a matched retouch.
- 01
Give the light a source
Specify what lights the face, what blocks it and where the shadow lands.
- 02
Separate the materials
Uneven skin highlights, stray hairs, matte cloth and small hardware reflections.
- 03
Place the camera
An extended arm, an ordinary setting and believable foreground perspective establish how the shot could happen.
Each reference has a job
The complete character sheet stayed in every generation. The selected night selfie supplied the photographic treatment.
Carry it into three lighting setups
The same identity references, with a new physical lighting situation for each scene.
Executed prompt excerpts & full prompt download
Reference roles: “INPUT1 complete canonical turnaround is mandatory for identity, outfit and entire body silhouette/proportions; INPUT2 accepted profile supplements body proportions; INPUT3 night selfie supplies the preferred photographic face, skin, hair texture and practical lighting realism.”
Daylight: “Open sky above/front-left provides broad daylight; shop awning blocks light on far cheek and underside of chin, strong believable difference between sky-facing planes and recessed shadows.”
Material behaviour: “Skin microtexture follows facial planes, not uniform grain stamped over face. Eyes aligned with a single consistent gaze, natural catchlights tied to real source.”
Download the three executed prompts · JSON ↓Review the result, then extend it
The accepted tests led to six more profile images: full figure, side view, expressions, seated contact, mirror logic and a walking turn.
Selection checks
- Recognisable face, silhouette and outfit
- Coherent eyes, fingers and small dental details
- Feet, hands and seated poses make physical contact
- Light direction agrees with shadows and reflections
What still needs judgment
The diner face remains a little polished; some flyaway hair sits near the frame edge. Prompts were not followed perfectly. These are visual comparisons across changed poses and settings, with no measured identity score.
“Phone-photo” describes the generated appearance. No actual camera capture, custom training or ComfyUI pipeline was used for this sequence.