We use cookies to improve your experience and analyze site traffic. By continuing to use this site, you agree to our Privacy Policy.

    Stable Diffusion Prompt Generator

    Free SDXL / SD 3 / Flux prompt builder with pre-filled negative prompts, sampler, CFG and steps.

    Why negative prompts matter more than positive prompts

    Stable Diffusion is famously unforgiving — without a negative prompt, even SDXL and SD 3 will routinely produce blurry textures, mangled hands, extra fingers, watermarks and that telltale "AI sheen." A well-crafted negative prompt blocks those failure modes before the model commits to them. Our generator pre-fills the negative prompt with the 14 most-cited tokens from the Stable Diffusion community (blurry, low quality, bad anatomy, bad hands, extra fingers, mutated, deformed, watermark, text, signature, jpeg artifacts, oversaturated, ugly, cropped). Tweak the list for your specific subject — for example, add "cartoon, anime, painting" when you want strict photorealism.

    Sampler, steps and CFG — the holy trinity

    • DPM++ 2M Karras + 30 steps + CFG 7 — the modern safe default for SDXL.
    • Euler a + 20 steps + CFG 6 — fastest, great for stylized art and bulk generation.
    • DPM++ SDE Karras + 40 steps + CFG 7 — slowest but most photorealistic skin and texture.
    • Flux.1 dev — uses 28–50 steps with CFG 1 (yes, one). Don't apply SDXL settings to Flux.

    Frequently asked questions

    What is a negative prompt in Stable Diffusion?

    A negative prompt tells Stable Diffusion what NOT to include. It's the single most powerful quality lever — adding things like 'blurry, bad anatomy, extra fingers, watermark, low quality' typically improves output more than any positive prompt tweak.

    What's the best negative prompt for realistic photos?

    For realism use: 'blurry, low quality, bad anatomy, bad hands, extra fingers, mutated, deformed, watermark, text, oversaturated, cartoon, anime, painting'. Our generator pre-fills this and lets you customize.

    What's the best Stable Diffusion sampler?

    DPM++ 2M Karras is the modern default — sharp results in 20–30 steps. Euler a is faster and great for stylized art. DPM++ SDE Karras produces the most photoreal output but is slower.

    What CFG scale should I use?

    CFG 7 is the safe default. Lower (4–6) for creative freedom and softer results. Higher (8–12) for strict prompt adherence but risks over-saturated, burnt-in artifacts above 12.

    Does this work for Flux as well?

    Yes — Flux.1 (dev and schnell) uses the same positive prompt structure but ignores negative prompts. Just leave the negative field blank when generating for Flux.