Make a glamour or editorial portrait
The look that used to need a booked studio, a light rig, and a photographer starts from a sentence now. Describe the shot — the subject, the styling, the light source, the lens feel — and the image models render it in up to 2K.
Prompt like a shot list, not a wish
The prompts that render commissioned-looking portraits are built like a photographer's brief: subject + styling + light + lens. "Editorial portrait, black satin dress, single soft key light, dark studio, catchlights in the eyes" is a shot list in one sentence.
Vague praise ("beautiful model, high quality") produces the stock-photo sameness everyone recognizes. The model already wants to make something pretty — your job is to tell it which pretty.
Prompts that work
"Glamour editorial portrait — off-shoulder evening gown, single key light, dark seamless backdrop, sharp catchlights." — the four-part shot list.
"Golden-hour balcony portrait, linen shirt, rim light, shallow depth of field, shot on 85mm." — lens language changes the render. 85mm compresses and blurs the background; 35mm keeps the environment in the story.
"Noir studio shot — hard side light, smoke haze, dramatic shadows." — one modifier for light quality, one for atmosphere.
"Beauty close-up — glossy lip, sculpted cheek light, clean skin texture, white backdrop." — tight crops respond to surface-level direction (gloss, texture, skin detail).
The words that move the render
Light is the highest-leverage word in a portrait prompt: key light, rim light, softbox, hard side light, backlight, window light. One lighting term changes the shot more than five adjectives about the face.
Backdrop next: "seamless", "dark studio", "environmental" (a real place behind them), "out of focus city". The background instruction decides whether it's a headshot or an editorial.
Styling words pin the look: garment + fabric + silhouette ("satin slip dress", "oversized wool coat"). Same trick as the outfit edit — the model renders texture it's been told about.
Camera terms are real: shallow depth of field, bokeh, 85mm, medium format, film grain. The model has seen enough photography that these behave like switches.
Keeping a face consistent
Text-to-image invents a fresh face every run — that's the feature and the limitation. When you land a face you like, that render becomes your base: future portraits are image edits on it (outfit, pose, light) rather than new rolls of the dice.
Describe the same person the same way each time — age, hair, features — and variation shrinks, though it never reaches true identity lock the way editing an existing image does.
Iterate cheap: prototype the look on the standard tier, and only when the prompt is proven do the final render on the 2K tier.
The honest limits
Hands remain the weak point of every image model — tight crops and poses with hands partly out of frame dodge the issue.
Renders land in your local library and auto-delete after 48 hours — download keepers when they happen.
Portraits of identifiable real people stay refused at the gate — this is for original characters.