WHAT NEW IN X10.30 FRI

AI collections from X · 22 collections · 61,986 posts

chatgpt-imagesTHREAD

GPT Image 2.5 prompt locks a face for a photorealistic character sheet

Upload one clear face photo and run a long prompt to generate a photorealistic three-view character sheet, emphasizing consistent features, film-flash texture and fully clothed styling.

Abdul Shakoor
@abxxaiArchived 2026.09.14Originally posted 2026.09.12
THREADPosted in 2 parts
  1. 1/ the woman character sheet made with GPT Image 2.5 in Arcads upload one clear photo of the face you want to lock, then run this prompt on it: use Sunburst for this one, not Flare. sheets are where you want the extra precision, everything after this depends on the face being right. Prompt: Create a photorealistic character reference sheet based on the woman in the image. Preserve her exact facial identity, structure, proportions and features. Identity & realism: Preserve her exact face shape, eyes, nose, lips, smile lines, brows and hairline from the image Sun-kissed tanned skin with a warm glow, visible pores, faint freckles across the nose and cheeks, natural oily sheen on the forehead, nose and cheekbones, slight redness, tiny natural imperfections, no over-smoothing Long light-brown hair with honey and sun-bleached golden tones, loose soft waves, slightly messy and humid from sea air, a few strands falling across her face, natural flyaways Real skin, real hair, no stylization, no CGI look, no illustration Pose & styling: Relaxed neutral standing pose, arms loose at her sides, not the laughing pose from the image Fully dressed at all times: black satin slip dress with thin spaghetti straps and a soft draped neckline, falling to mid-calf; a fine thin gold chain necklace with a tiny pendant; small gold huggie hoop earrings; bare hands, no rings Expression calm with a soft closed-mouth smile, except in the laugh panel Look (same in every panel): Shot like a late 1990s 35mm point-and-shoot film camera with direct on-camera flash: slightly hot frontal flash on the skin, soft focus, heavy fine film grain, faded warm color with a slight magenta-green cast, lifted soft blacks, gentle halation on highlights Background: plain warm off-white paper backdrop, flat and even, no environment, only a soft flash shadow directly behind her Layout: Title at the top: "LENA — CHARACTER SHEET" Top row, evenly spaced: Full Body Front, Full Body Side, Full Body Back Bottom row, evenly spaced: Front Face Close-up, Three-Quarter Face Close-up, Left Profile Close-up, Laugh Close-up (wide open laugh with teeth, eyes crinkled, exactly like the image), Hair Detail Close-up, Shoulder Close-up Design details: Each frame enclosed in thin clean borders of consistent thickness Above each frame a centered readable serif label naming the view Labels never overlap the images Final quality: ultra-detailed real film photography, identical face, hair and outfit in every panel, consistent flash and color across all panels, clean symmetrical professional layout, no distortion, no missing details

    GPT Image 2.5 prompt locks a face for a photorealistic character sheet (thread part 1)
  2. 4/ the location plate made with GPT Image 2.5 (Sunburst) this is the part everyone skips. you don't describe the harbor to the video model, you build the actual table first and hand it over as frame zero. Prompt: Create a photorealistic environment reference sheet of the location in the image, with no people anywhere in any panel. Location: a small outdoor trattoria terrace at the water's edge of a pastel Italian fishing harbor village, a late summer evening in 1998. Lighting lock, identical in every panel: deep blue hour dusk sky, warm bulb string lights hanging overhead, an old iron street lamp glowing amber, harbor lights reflecting in long gold streaks on dark calm water, the table area lit by a hard frontal on-camera flash while the background falls into blue dusk. Panels: Top, full width: Wide Establishing. The terrace seen from along the quay: a row of small tables with red and white checkered tablecloths and wooden cane chairs, a cream canvas umbrella printed with "TRATTORIA" in dark serif letters, warmly lit ochre, pink and terracotta houses stacked up the hill, moored white fishing boats and sailboat masts, a dark tree-covered hillside against the dusk sky. Bottom left: Medium Table View. The corner table at the water's edge: red and white checkered cloth, a white plate of spaghetti in tomato sauce with a fork resting in it, a tumbler of cloudy lemonade with a lemon slice, a dark green glass bottle, two water glasses, a crumpled white napkin, two wooden cane chairs facing each other, the harbor and boats behind. Bottom right: Close Detail. The tabletop up close: the checkered cotton weave, condensation on the lemonade glass, the lemon slice, sauce on the pasta, the fork, the flash glare on the plate rim. Look, same in every panel: late 1990s 35mm point-and-shoot flash photograph, heavy fine grain, soft focus, faded warm color with a slight magenta-green cast, lifted soft blacks, gentle halation on the string lights. Layout: title at the top "HARBOR TRATTORIA — LOCATION PLATE". Each panel in thin clean borders with a centered readable serif label above it. Labels never overlap the images. Final quality: ultra-detailed real film photography, the same place, same objects and same light in every panel, no people, no distortion.

    GPT Image 2.5 prompt locks a face for a photorealistic character sheet (thread part 2)
Abdul Shakoor @abxxaiOriginally posted 2026.09.12
Open original on X
x.com/abxxai/status/2098788823677514052Archived 2026.09.14 19:43