Prompt guide
How to write AI porn prompts that render what you meant
Most disappointing AI porn images come from a good idea described badly. This guide gives you a repeatable structure, the vocabulary that models respond to, and twelve templates you can paste straight into the prompt lab or the studio.
1. The eight-block anatomy
Diffusion models weight tokens roughly in the order they appear and treat commas as soft separators. Write in blocks, most important first:
| Block | What it decides | Example tokens |
|---|---|---|
| Subject | Age (adult), build, one identity anchor | stunning 25yo woman, athletic 28yo woman |
| Look | Hair, eyes, skin details | auburn curls, freckles and green eyes |
| Outfit | Fabric and coverage — fabric words carry texture | black lace lingerie, sheer white robe |
| Pose | Body position and relation to furniture | reclining on pillows, kneeling on the bed |
| Setting | Location and time of day | dim hotel bedroom, rainy night penthouse |
| Lighting | The single biggest realism lever | moody chiaroscuro, golden hour backlight |
| Camera | Focal length, depth of field, film stock | 85mm, shallow depth of field, Kodak Portra look |
| Finish | Render quality and grade | ultra detailed skin, 8k, cinematic color grade |
A complete prompt following this order:
2. Lighting vocabulary that actually changes the render
- Soft window light at dusk — low contrast, cool shadows, flattering on skin. Default for boudoir.
- Moody chiaroscuro — one hard source, deep shadows; adds drama and hides background clutter.
- Warm candlelight glow — orange cast, soft falloff; pairs with
film grain. - Golden hour backlight — rim-lit hair, haze; ask for
lens flareexplicitly if you want it. - Neon rim light — saturated edges; name the colours (
cyan and magenta) or the model picks its own. - Single bedside lamp — practical light in frame, intimate scale.
3. Camera tokens and what they imply
- 85mm portrait lens — compressed features, creamy background. The safest realism token.
- 35mm environmental — shows more room; slightly wider faces at close range, so step back with
full body. - Shallow depth of field — isolates the subject, softens hands and feet (useful when they misbehave).
- Film grain, Kodak Portra look — breaks the plastic digital sheen; Portra implies warm, muted skin tones.
- RAW photo, sharp focus — pushes toward unretouched realism; combine with
natural skin texture.
4. Preset-specific adjustments
Photoreal
Keep camera and lighting blocks. Avoid masterpiece, best quality and stacked superlatives — they add nothing on modern models and crowd out real information. Say natural skin texture instead of perfect skin.
Anime
Replace camera tokens with art direction: cel shading, clean line art, detailed eyes, glossy highlights. Name the composition (full body, bust shot) because there is no lens to imply it.
Painterly
Lead with medium: oil on canvas, watercolour wash, loose brushwork. Palette words (warm palette, pastel) matter more than lighting words. Painterly is the most forgiving preset for hands.
5. Aspect ratio is part of the prompt
The studio offers 15 ratios for images and five for video. Practical defaults: 2:3 for standing or reclining full-body shots, 1:1 for busts and avatars, 16:9 for scenes where the setting matters, 9:16 for phone wallpapers and vertical video. A pose that fights the frame (a lying pose in 9:16) is the most common cause of cropped heads and extra limbs.
6. Negative prompts, used sparingly
Where a negative field is available, keep it short and structural: extra fingers, deformed hands, blurry, watermark, text. Long negative lists fight the positive prompt and flatten lighting. If a problem persists, fix it in the positive prompt first (hands resting on thighs) before adding negatives.
7. Iterating: change one token at a time
- Render the template as-is.
- Change only the lighting block. Compare.
- Change only the camera block. Compare.
- Now change subject or outfit. You will know exactly which token moved what.
Once you like a still, extend it instead of re-rolling: upscale to 2K–8K for print-grade detail, edit the background or clothing with a plain-English instruction while naming what must stay unchanged, or animate the still into a 2–30 second clip with a motion prompt (see templates 11 and 12).
8. What not to write
Prompts describing anyone under 18, non-consent, incapacitation or family scenarios are blocked at submission. Real people cannot be depicted or face-swapped without documented consent, and the face swap tools ask for that confirmation on every job. Fictional adults, always.