If you are using GPT Image 2 for Hong Kong–style street portraits with a cinematic look, the fastest way to get consistent results is not “more adjectives”—it is a repeatable prompt structure you can share, version, and debug.
This guide gives you a modular GPT Image 2 prompt blueprint for cinematic Hong Kong retro street fashion portraits. Copy the blocks in order, replace the <placeholders>, and keep the “must-lock” constraints (identity, wardrobe, lens language) separate from the “creative” parts (neon layers, motion blur). A short FAQ at the end answers the most common search intents: modular prompts, identity drift, and busy backgrounds.
Who this template is for
- Full-body or half-body street shots that should feel like editorial or film stills, not studio glamour retouching.
- Teams that need stable identity (bone structure, believable skin texture) while still getting atmosphere from the street.
- Creators building prompt libraries for social covers, portfolios, or lookbook inspiration.
What “cinematic + retro Hong Kong” means in prompt language
Set expectations in a few concrete phrases: layered neon signage, wet pavement reflections, warm film lighting, shallow depth of field, and foreground motion blur for a candid fashion feel. The modular sections below map directly to those goals.
Modular prompt blocks (concatenate in order)
Concatenate the following sections into one prompt. Replace anything in <angle brackets> with your own scene, wardrobe, or pose notes.
Theme
Hyper-real cinematic portrait, high-end fashion, 8K quality, magazine-style photography. Set aspect ratio as needed (for example <3:2> or <16:9>).
Subject
Use reference image 1 as the primary subject only. Match facial structure and natural skin texture (visible pores). Keep makeup consistent with the reference. Do not change ethnicity or gender.
Hair
Match hairstyle and color from reference image 1. Wind-blown strands across the face in a stylish way, partially obscuring features.
Makeup
| Step | Direction |
|---|---|
| Base | Clean, breathable, soft matte with subtle luminosity |
| Eyes | Natural enlargement, long lashes, clear gaze |
| Lips | Glossy “juicy” lips, soft pink-to-peach gradient, gentle peaks |
Skin: hyper-real texture, visible pores, natural micro-contrast—avoid plastic skin.
Wardrobe
Match the upper-body outfit from reference image 1. Add detail in <wardrobe notes> (materials, sheen, accessories).
Scene / environment
<Upload a scene reference and describe it>: for example layered neon signage, wet pavement reflections, or soft overcast alleys—keep the palette aligned with retro Hong Kong street mood.
Action / motion
Foreground pedestrians in fast motion with motion blur (long-exposure feel) crossing the lens, partially blocking the frame. <Pose reference or text>: editorial model posture; subject positioned on the right.
Composition
Low-angle looking-up camera; full-body horizontal portrait; subject biased to the right; clean lines and restrained negative space.
Camera / lens
24mm wide perspective; shallow depth of field; cinematic bokeh; professional fashion-realism rendering.
Lighting
Soft cinematic warm light with clear mood; controlled highlights; shadows retain detail.
Color / finishing
High detail, premium color grading, natural skin texture, subtle film character (no obvious grain). No text, logos, or watermarks.
Output
Set resolution and aspect ratio for your platform (for example <2K HD> or a specific long edge).
Image note (different framing from the cover)
The example below is in addition to the cover image at the top—use it as a second visual anchor alongside the prompt blocks.
FAQ
Why modular prompts for GPT Image 2?
Street scenes add many variables. Modules separate “must-lock” constraints (face, wardrobe) from “creative” elements (pedestrians, neon density), which makes failures easier to debug.
Identity drift—what should I tighten first?
Strengthen the reference image 1 block (bone structure, pores, makeup match) and keep the rule do not change ethnicity or gender. Remove conflicting “flawless skin” wording if it fights texture cues.
Background too dominant?
Increase subject separation / bokeh cues, or reduce saturated signage and extreme contrast in the scene block so the subject remains the visual anchor.
Quick troubleshooting checklist
- Over-smoothed skin: repeat visible pores and natural texture; avoid “flawless” phrasing.
- Too much motion blur: switch to subtle motion blur and limit how much of the frame pedestrians cover.
Save this layout as a master template and only swap the scene / wardrobe / pose paragraphs to scale a repeatable GPT Image 2 portrait pipeline.
