What is an AI image prompt?
An AI image prompt controls what an image model puts in frame and how it renders. It names the subject, the style, and the details you want — a character, a wardrobe, a lighting setup, a camera angle — and how you phrase each of those layers steers the output. An AI video prompt does the same for motion, often driving an image-to-video base like Wan 2.2 from a starting frame.
A prompt is not the whole story. The same words produce wildly different output depending on the generation settings behind them: the sampler, the number of steps, the CFG scale, the seed, and the output dimensions. That is why every entry here publishes those parameters alongside the text, and includes the negative prompt whenever the original creator used one.
These are real prompts pulled from real generations, not invented examples. Each one is attributed to the creator who made it on Civitai, so the prompt you copy is the prompt that produced the image you see. Generic prompt lists are usually built around tools that refuse adult subject matter, so their prompts quietly assume a filtered model. Everything here is built for uncensored and unrestricted output, so NSFW prompts work as written.
Why this prompt library beats a generic prompt list
Most prompt collections give you a paragraph of text and nothing else, so when your result differs from the example you have no way to know whether the sampler, the step count, or the seed was responsible. Here every parameter is published, and the four things that set this archive apart are visible on every prompt page.
- Full reproducible settings. Each prompt ships with sampler, steps, CFG scale, seed, and dimensions — the parameters that actually determine output, not just the words.
- Real results, not mockups. The preview on each card is what that exact prompt produced.
- Cross-model gallery. See the same prompt rendered across multiple models side by side before you spend render time, so you pick the right base model first.
- Honest attribution. Every prompt is credited to its original creator and sourced from Civitai, so you can trace where it came from.
- Built for uncensored models. Safety levels run from SFW through Soft and NSFW to Unrestricted, so adult prompts are labeled and findable rather than buried.
Anatomy of a great AI image prompt
Strong AI image prompts are built from layers, not a single sentence. Each layer steers a different part of the render, and the most reusable entries make every layer explicit. Use these six parts as a checklist when you write or tweak a prompt:
- Subject — who or what is in frame. The core noun: a person, a creature, an object. A concrete subject gives the model less room to drift, so results stay closer to your intent.
- Descriptors — the adjectives that define the subject: body type, age presentation, expression, pose, and explicitness. For NSFW prompts this is where the uncensored detail lives.
- Wardrobe & props — clothing, lingerie, accessories, or the absence of them, plus any objects in the scene. Specifying material and fit (latex, sheer, wet) sharpens realism.
- Lighting — the single biggest lever for mood and realism: soft window light, hard rim light, neon, golden hour, studio softbox. Lighting terms separate flat output from photographic output.
- Camera & lens — focal length, aperture, and shot type: 85mm portrait, f/1.4 shallow depth of field, low-angle, full-body. These cues push diffusion models toward photographic framing.
- Composition & style — the overall look: photorealistic, anime, painterly, film grain, aspect ratio, and any model-specific style tokens that a checkpoint like Pony or Illustrious responds to.
How metadata changes the output: sampler, steps, CFG, seed
Identical prompt text can produce wildly different images depending on four settings. Understanding them is the fastest way to stop fighting your results, and every prompt here lists all of them so you can match the source exactly.
Sampler is the algorithm that turns noise into an image. Different samplers converge at different speeds and with different texture; swapping the sampler on the same prompt changes detail and sharpness, not just quality.
Steps is how many denoising passes the model runs. Too few and the image stays soft and underdeveloped; past a point, extra steps add render time without adding detail. The right number is sampler-dependent, which is why a copied step count only makes sense alongside the sampler it was tuned for.
CFG scale (classifier-free guidance) controls how literally the model obeys the prompt. Low CFG gives the model creative latitude and softer results; high CFG forces strict adherence but can over-saturate or burn detail. Most checkpoints have a recommended CFG range, which is why copying the source value is safer than guessing.
Seed is the starting noise pattern. Reuse the exact seed with the same prompt and settings and you reproduce the image; change the seed and you get a fresh variation of the same idea. Locking the seed is how you iterate on wardrobe or lighting without losing a composition you like.
Carry sampler, steps, CFG, seed, dimensions, and the negative prompt together and you can reproduce a result first, then change one variable at a time to learn what each does. A prompt copied without its settings is half a recipe.
How to use these AI prompts
Every prompt here is free to copy and reuse. The workflow is the same whether you generate locally or in the cloud:
- Browse or filter — narrow the grid by model, by kind (image or video), by safety level from SFW to Unrestricted, and by rating. Sort by popular, quality, or newest.
- Open a prompt — read the full prompt text, the negative prompt when present, and the complete settings: sampler, steps, CFG scale, seed, dimensions, and video duration.
- Copy the prompt — take the exact text and paste it into a local install of the model, or run it instantly on ZenCreator, a hosted generator that needs no GPU and no install.
- Check the cross-model gallery — open the gallery view to see the same prompt rendered across base models, so you can pick the one that renders your subject best.
- Tweak and re-run — lock the seed and change one element at a time. Swap the lighting or lower the CFG to keep the composition while you adjust the subject.
Uncensored and NSFW prompts, written to work
The angle here is uncensored output. Prompts are labeled by safety level — SFW, Soft, NSFW, and Unrestricted — and the explicit ones are stored and copied exactly as the creator wrote them, with the same full metadata and attribution as every other entry.
Because the underlying models are open-weight checkpoints like Flux, SDXL, Pony, and Illustrious, the prompt is the only thing standing between you and the result, with no platform-side content rewrite in the way. Filter to the NSFW or Unrestricted levels to see only adult prompts, then copy and run them locally or on ZenCreator.
AI prompts for every model
Prompts are not universal — phrasing that excels on Flux can fall flat on Pony, and Pony's style tokens mean nothing to SDXL. Each entry is filed under the model that actually produced it, so you copy prompts that are already tuned to the checkpoint you run.
On the image side you will find Flux and Flux Dev prompts, Pony Diffusion XL prompts, SDXL prompts, Illustrious prompts, and CyberRealistic prompts, among others. For video, AI video prompts cover image-to-video bases such as Wan 2.2, including the duration each clip was rendered at, so the prompt is written for motion rather than describing a still frame. Filter to a single model to see only the prompts proven to work there.
