AI Prompts
Stable Diffusion prompts.
Prompt formulas, style keywords, negative prompts and settings for Stable Diffusion — tap to copy, then paste into your generator.
Tap Copy on any prompt, paste it into Stable Diffusion (Automatic1111, ComfyUI, Forge or similar), and replace the [bracketed] parts with your own subject.
Prompt structure
The core formula
[subject], [style / medium], [lighting], [camera / composition], [quality]. Example: a red fox curled in autumn leaves, digital painting, soft golden-hour light, close-up shot, highly detailed, sharp focus.
Add camera & lens detail
[subject], shot on [camera — e.g. Sony A7 III], [lens — e.g. 85mm f/1.4], shallow depth of field, bokeh, natural lighting, ultra realistic.
Quality boosters
[your subject and style], masterpiece, best quality, highly detailed, sharp focus, 8k, intricate detail, professional. Add these at the end — but don't overload; 3–5 tags is plenty.
Photorealistic
Portrait
RAW photo, portrait of [person — e.g. a woman in her 30s], detailed skin texture, freckles, soft window light, 85mm lens, shallow depth of field, photorealistic, high detail.
Product shot
professional product photo of [product], on a clean white seamless background, studio softbox lighting, sharp focus, high detail, commercial photography, 4k.
Landscape
landscape photo of [place — e.g. a misty mountain valley at dawn], dramatic sky, golden hour, wide-angle, ultra detailed, high dynamic range, National Geographic style.
Art & illustration styles
Anime
[subject], anime style, cel shading, clean line art, vibrant colours, detailed eyes, studio-quality illustration, by [artist / studio influence].
Oil painting
[subject], oil painting, thick impasto brushstrokes, rich colour palette, dramatic chiaroscuro lighting, textured canvas, classical fine art.
3D render
[subject], 3D render, Octane render, Pixar style, soft global illumination, subsurface scattering, smooth materials, high detail, 4k.
Pixel art
[subject], pixel art, 16-bit, retro game sprite, limited colour palette, crisp pixels, isometric view. Generate small (e.g. 512px) for cleaner pixels.
Negative prompts
Universal negative
lowres, blurry, jpeg artifacts, watermark, text, signature, out of frame, cropped, worst quality, low quality, ugly, duplicate. Paste this into the negative prompt box.
People & hands
deformed hands, extra fingers, missing fingers, mutated hands, bad anatomy, disfigured, extra limbs, fused fingers, malformed, poorly drawn face, asymmetrical eyes.
Keep it clean & realistic
cartoon, anime, illustration, painting, 3d render, cgi, plastic skin, oversaturated, airbrushed. Use this to push a prompt toward a realistic photo look.
Settings & parameters
Reliable baseline
Sampler: DPM++ 2M Karras · Steps: 25–30 · CFG scale: 7 · Size: 512×512 (SD 1.5) or 1024×1024 (SDXL). A safe starting point for almost any prompt.
Choosing a sampler
DPM++ 2M Karras: sharp, dependable all-rounder. Euler a: fast, softer, more painterly. DPM++ SDE Karras: extra detail, a little slower. Keep the seed fixed to compare fairly.
Resolution & hi-res fix
Generate at the model's native size (512 for SD 1.5, 1024 for SDXL), then enable Hires. fix with upscale 1.5–2×, denoising 0.3–0.5. Generating too large upfront causes duplicated bodies.
Lock a look with seed
Set Seed to -1 to explore. When you get a composition you like, copy its seed and reuse it while you tweak the prompt or settings so the base image stays consistent.
Advanced
Attention weighting
Wrap a term in parentheses with a weight to emphasise it: (red scarf:1.3) makes it stronger, (background:0.6) makes it weaker. Stay roughly between 0.5 and 1.5 to avoid artifacts.
img2img from a reference
Upload a base image in img2img, keep your prompt, and set Denoising strength: 0.4–0.6 to restyle while preserving the composition (lower = closer to the original, higher = more freedom).
Inpainting a fix
Mask only the area to change, prompt just that region (e.g. [a straw hat]), set "Inpaint masked" with "Only masked" content, denoising 0.5–0.75. Great for fixing hands, faces or swapping details.
How to prompt Stable Diffusion
Stable Diffusion reads your prompt as a bag of weighted keywords rather than a sentence, so front-load the important terms and separate ideas with commas. Build from the core formula — subject, style, lighting, camera, quality — then refine with a good negative prompt to remove the usual failures like blurry textures and deformed hands. When something is nearly right, lock the seed and change one thing at a time; that's how you learn what each keyword and setting actually does. And remember the settings matter as much as the words: sampler, steps and CFG scale can turn the same prompt into a very different image.
FAQ
What is a negative prompt?
It's a list of things you don't want in the image — like blurry, deformed hands, extra fingers, watermark or lowres. Stable Diffusion steers away from those terms, so a good negative prompt is one of the fastest ways to clean up your results.
What CFG scale and steps should I use?
Start around CFG 7 and 25–30 steps for most images. Lower CFG (4–6) gives the model more creative freedom; higher CFG (9–12) sticks closer to your prompt but can look over-cooked. More steps rarely help beyond 30–40 with modern samplers.
Which sampler should I pick?
DPM++ 2M Karras is a reliable all-rounder that looks good at 20–30 steps. Euler a is fast and a little more painterly. Try the same prompt and seed across a couple of samplers to see which suits the look you want.