Cinematic AI Image Prompts and Character Lock
Paste-ready photoreal prompts for Nano Banana Pro, GPT Image, and Higgsfield Soul, plus character sheets and face-locked outfit swaps.
Before & after
Same prompt, same agent — the only difference is the skill.
Here's a cinematic image prompt for a lighthouse at dusk: "A lone lighthouse standing on a rocky cliff at dusk, dramatic sky with orange and purple hues, waves crashing against the rocks below, moody atmosphere, cinematic lighting, highly detailed, 4k, professional photography" You could enhance it by specifying: - A camera angle (wide shot, aerial view) - A mood or color palette - Weather conditions - The style of a specific photographer or film Want me to write a few variations with different moods?
Cinematic prompt in a hyper-real stack — real camera vocabulary, not
adjective soup. "4k, highly detailed, professional" is what produces
slop; specific optics is what produces a photograph.
READY TO PASTE — Nano Banana Pro / GPT Image
"Lone lighthouse on a granite headland, blue hour, 8 minutes after
sunset. Shot on a 35mm lens at f/4, medium-wide, slight low angle from
the wet rocks. Key light: the last warm sun raking from camera-left
across weathered white paint, cool ambient fill from the deepening sky.
Practical: the lighthouse lamp just lit, warm sodium against blue.
Texture: salt-crusted stone, paint chalking at the seams, fine sea mist
softening the far edge. Long exposure — water goes to silk, one gull
motion-blurred. Muted teal-and-amber grade, deep but not crushed
shadows, natural film grain. Photographic, not illustrated."
WHY EACH PART EARNS ITS PLACE
35mm f/4 → the depth and compression of a real photo, not
the flat everything-sharp look that reads as AI
blue hour + 8 min → a specific light, so you get one look, not an
averaged "dusk"
named textures → grain/pores/chalking is what sells hyper-real;
"highly detailed" does nothing
practical lamp → a real light source in frame grounds the whole
image
BEFORE YOU SPEND CREDITS
Run the pre-prompt checklist: lock lens, hour, key-light direction, and
grade first. Generating variations before those are fixed is how you
burn a credit pack chasing a look you never defined.
Want three graded variations (storm / clear / fog), or a character sheet
if a keeper is in frame?About this skill
Your AI portraits look plastic, your character's face changes every shot, and "cinematic, 8k, masterpiece" keeps giving you glossy stock photos. This skill writes image prompts the way a cinematographer writes a shot list. It specifies the focal length, aperture, the motivated light source and its direction, one film stock, and the exact skin imperfections that make a face look real. It also locks your character with a fixed text block and reference sheets, so shot five shows the same person as shot one.
What it does
- Rewrites vague or adjective-heavy prompts into ordered, paste-ready paragraphs: camera, subject, wardrobe, environment, light, lens/stock/grain
- Diagnoses plastic skin and flat lighting in your existing prompt and shows before -> after fixes
- Builds a reusable identity block plus 3-panel face and full-body character sheet prompts on a neutral gray backdrop
- Writes outfit-swap edits that change only the clothing and leave the face, pose, and light alone
- Labels each attached reference with its role so the model stops blending faces with garments
- Recommends Nano Banana Pro, GPT Image, or Higgsfield Soul based on what the shot needs, and flags anything a given model can't do
What you get
Prompts in code blocks, ready to paste, with model, aspect ratio, and reference roles at the top, plus a short note on why each choice matters and which single change to try next.
Best for
AI filmmakers, music-video and ad-creative teams, and creators who need one recurring character to stay consistent across a series of stills or keyframes.
SKILL.md preview
---
name: cinematic-ai-image-prompts-character-lock
description: Writes paste-ready photoreal image prompts for Nano Banana Pro, GPT Image, and Higgsfield Soul using real lens, light, and film-stock vocabulary, plus identity blocks, 3-panel character sheets, and outfit-swap edits that keep a face locked across shots. Use when the user says "write me a Nano Banana prompt", "make this look like a real photo", "my character's face keeps changing between shots", "build a character sheet", "swap the outfit but keep the face", "my AI portraits look plastic", "which image model should I use for this shot".
---
# Cinematic AI Image Prompts and Character Lock
… (sign up to view the full skill)More video & editing skills
View all Video & Editing skills →Character and Location Reference Sheets for AI Video
Locks faces, wardrobe and sets before AI video: identity blocks, 3-view turnarounds, empty location plates, and reference role lines.
AI Image Realism Pass: Fix Plastic Skin and Flat Depth
Diagnoses why AI characters look plastic or pasted-on and rewrites prompts with gray reference sheets, lit haze, real optics, and depth.
AI Visual Style Reference: Style Card and Drift Test
Turns a look into a reusable AI style reference: measured style card, paste-in style block, reference-image plan, and a drift test.
AI Image Prompt Doctor: Lighting, Lens and Fixes
Diagnose why AI images come out generic or wrong and rewrite the prompt with named light, lens, framing and model-correct syntax.