NSFW AI Prompts: How to Write Prompts for Each Spicy Model

Write the prompt as one to three plain sentences, in this order:

1. Subject: an adult character with an age (“a man in his thirties”), build, face, hair and outfit.
2. Setting: a specific place and time, such as a penthouse at dusk or a cabin in winter.
3. Light: one clear light source: candlelight, neon, window light or golden hour.
4. Camera: framing and lens, like a close-up or a full-body shot, 35 mm or 85 mm, shallow depth of field.
5. Style: realistic photo, film grain, cel-shaded anime or oil painting.
6. Mood: two or three words, such as “sultry and calm” or “playful and warm”.

These models read plain sentences better than long tag lists, and Stable Diffusion weighting such as (word:1.4) does nothing useful here. If a word doesn’t change the picture, cut it.

Spicy UNCENSORED follows long, specific prompts well and is the model for finals at 1K or 2K. A prompt is required, even when you upload references.

For edits, attach as many as 10 reference images and name them by upload order: Figure 1 for the first, Figure 2 for the second. Spell out what stays and what changes: “Keep the face and hair from Figure 1, use the outfit from Figure 2, and place her on a rooftop at night.”

• Pick the aspect ratio for the shot: 2:3 or 3:4 for portraits, 16:9 or 21:9 for wide scenes, 9:16 for phone screens.
• Draft at 1K, then run the winner again at 2K with the same prompt.

Wan 2.7 Image charges a flat 6 credits whether you attach no reference or 9, so it is the place to explore. It keeps the structure of a reference image well, which helps when you want to hold a pose or a composition and change the style or the setting.

Its aspect ratios are 16:9, 9:16, 4:5, 4:3 and 1:1. Write edits as short instructions: “Change the background to a beach at sunset and keep everything else.”

Prompt expansion is off, so Wan 3.0 follows your words as they are and adds nothing. Describe the clip in full: the subject, one main action, the camera move, the light, the sound and the pace.

For text-to-video, describe the scene as well as the motion. With a first frame, describe only what changes; the picture already shows the rest.

• Put the sound in the prompt: “rain on the window, a train passing in the distance”.
• Write long takes in beats: “she walks to the window, pauses, then turns back toward the camera”.
• Name camera moves plainly: slow push-in, pull-back, gentle orbit, static shot, handheld.

HappyHorse 1.1 animates a first frame. You can skip the prompt or write as many as 2,500 characters. Start with only the motion and the camera, like “slow push-in, she smiles and looks away”, and add detail only if the result drifts.

Because the clip stays close to the image, fix the composition, the outfit and the light in the still before you animate it.

Swap the parts in square brackets and keep the order:

• Portrait: “Portrait of [adult character with age, hair and outfit] in [setting], [light source], 85 mm lens, shallow depth of field, [style], [mood].”
• Scene: “[Adult character] [doing one thing] in [place] at [time of day], wide shot, [light], [style], [mood].”
• Edit: “Keep the face and hair from Figure 1. Change [the one thing]. Keep the light and framing.”
• Video: “[Adult character] [one action], [camera move], [light], [sound], [pace].”

• The character doesn’t read as clearly adult: write an adult age and adult features (“in her thirties”, “mature features”) and cut vague, soft words. If a result still looks under 18, delete it and don’t use it.
• Skin looks plastic: ask for “natural skin texture”, “film grain” or a real lens, and stop stacking words like “perfect” and “flawless”.
• The face changes between images: use a reference image as Figure 1 and change one thing per run.
• The video ignores part of the prompt: cut it to one main action and move the rest of the detail into the first frame.
• A prompt is rejected: every prompt passes an automated safety filter. Rewrite it within the Terms of Service; Spicy Studio doesn’t publish workarounds.