These textures rely on consistent repeating structure; when the image is resized down, those patterns alias into moiré or noisy blobs. If you need them to survive, generate at higher resolution, avoid extreme sharpening, and keep the pattern region large enough in the frame. Post-process with careful upscaling rather than forcing more “detail” words, because word-driven micro-detail often becomes unstable noise when resized.

Use: “bad perspective, warped perspective, foreshortening errors, disproportionate limbs”. Also avoid prompts that imply extreme lens distortion. Add a positive anchor like “normal perspective, balanced proportions”.

Often harsh sharpening/contrast stacks, or overly stylized makeup cues. For photoreal lips, use “natural lips, natural texture” and soft lighting. Avoid forcing “glossy lipstick” unless you want it. If lips smear, reduce CFG slightly and remove hyper detail words.

POSITIVE: RAW smartphone selfie, adult woman, black hoodie, convenience store at night, bright cool lighting, snack aisle, arm extended, (sharp eyes:1.2), crisp edges, no readable text NEGATIVE:

Sometimes, but composition is a bit more stubborn than small objects. Still, you can nudge it: (centered composition:1.15), (waist-up portrait:1.2), (head and shoulders:1.2). The bigger wins come from removing conflicting shot types and keeping the prompt simple. If you ask for “full body” plus “tight portrait,” the model will compromise. For consistent composition, lock a baseline phrase (“waist-up portrait photo”) and keep it unchanged while you iterate other details.

POSITIVE: RAW smartphone selfie, adult woman, casual jacket, car selfie, sunroof daylight creating soft top light, arm extended holding phone, (sharp eyes:1.2), clean edges, natural tones NEGATIVE:

Yes—disk or database pressure can cause intermittent save issues. The generator might be fine while storage lags.

We support privacy tools like VPNs and proxies. But those tools can also be used for abuse, so security systems may still react to risky patterns to keep the platform safe.

They can. Any text you submit may be checked for policy issues. Avoid writing prohibited terms even “as negatives” or “as a joke.”

Winks are tricky. Use “subtle wink” and keep the face large and well-lit. If it deforms, inpaint the eye region with a tight mask and prompt only “subtle wink, natural eyelids.”

Include your username, the email on the account (if you have access), and a rough timestamp. If payment is involved, include the invoice/receipt ID (do not post full card details).

Use a proven set recipe: “studio portrait photo, seamless backdrop, softbox key light, gentle fill, 85mm, shallow depth of field, subject centered.” Keep the outfit simple, keep expression simple, and don’t add busy environment cues. This setup is forgiving: clean background, controlled light, stable lens. Once it looks real, you can add small variations (colored backdrop, rim light) without collapsing into AI chaos.

Use a basic artifact block: “glitch, artifact, broken edges, jagged edges, aliasing”. Then reduce the common causes: oversharpening words, too much micro-detail stacking, and high-contrast lighting. Hair and fingers are high-frequency areas; they break first when the prompt is overloaded. Use clean lighting and keep the background simple so edges separate naturally.

Use negatives that describe the processing artifact: “oversharpened, overprocessed, crunchy, clarity filter, harsh sharpening”. Don’t negate “sharp focus” globally unless it’s truly the problem—often you still want a sharp subject. Also avoid stacking “sharp focus + ultra-detailed + high contrast” in the positive prompt; that combo commonly causes the crunchy look. If it persists, reduce steps slightly or keep CFG in the 5–7 range before adding more negatives.

Do an A/B test. Run the exact same prompt/seed/settings twice: once with your baseline negative, once with a minimal negative (just “text, watermark, logo” and maybe “bad anatomy”). If the minimal negative looks better, your negative list is likely suppressing something useful or contains contradictions. Then re-add your negative blocks one at a time until you find the offender. This is the fastest way to identify a bad negative term without guessing.

Not automatically. But frequent IP changes can look like account sharing or automation if combined with other signals. If you travel a lot or switch networks often, that can increase “false alarms.”