It competes with everything else that wants the same “visual budget.” If you weight “(cat ears:1.4)” and also describe long hair, a hat, bangs, headphones, and a busy background, the model has to allocate attention across all those head-area concepts. That’s why ears vanish or morph—your head is overloaded with competing items. The fix is often to simplify the area: remove hat/headphones, reduce hair complexity, or move background detail out. Weights don’t create infinite capacity; they just tell the model what to prioritize when capacity is limited.

Because shared IP ranges can have higher abuse rates. Security systems may add friction to reduce attacks and protect accounts.

It’s a great hand-saver pose in casual outfits. If pockets look weird, use “jacket” or “hoodie” outfits where pockets are easier to read.

If a job is stuck, the server may have restarted or the queue entry didn’t update. Refresh once; if it still shows queued after a long time, try generating again. If it happens often, open a ticket with the time and your image_id if you have it.

If your device blocks cookies or uses a privacy/cleaner tool, the session can’t persist. Make sure cookies are enabled for the site and disable any “block all cookies” setting. If you’re using a VPN/proxy, try without it—some networks trigger extra security checks.

Some checkpoints default to glam looks. Add “no makeup, minimal makeup” early in the prompt, and remove any style presets that inject makeup tags. If it’s stubborn, add negatives like “heavy makeup, lipstick, eyeliner” (but don’t overdo it or you can distort the face).

Models love adding “detail candy.” Fight it by declaring simplicity early: “no jewelry, no tattoos, simple outfit, clean look.” Then reinforce in negatives: “jewelry, necklace, earrings, tattoo.” If you *do* want one accessory, specify only that one: “small silver stud earrings” and optionally add “no other jewelry.” This “allow list” approach is more reliable than trying to subtract after the fact.

Some galleries only index certain folders. If images are in Downloads, your Gallery may not scan it immediately. Open the file from Downloads or move images to Pictures/DCIM.

Brands are learned as associations with certain aesthetics, lenses, and post-processing. Sometimes they help (Canon portrait look), sometimes they pull in irrelevant training artifacts. If you want consistency, avoid brand names and stick to physical descriptors: “85mm, f/1.8, ISO 200, natural light.” If you do use brands, keep it to one and don’t mix with other camera ecosystems.

Those are different visual priors: photoreal models expect real lighting and texture, while anime shading expects simplified gradients and stylized edges. Mixing them often yields plastic skin with drawn outlines because the model tries to satisfy both at once. If you want a controlled hybrid, pick a primary style (photo OR anime) and add only one small secondary cue (e.g., “slightly stylized color grading”) rather than full-on conflicting style families.

Hands improve when pose and framing are simpler. In the wizard, avoid extreme poses (hands toward camera, fingers splayed) until later. Use negative helpers like “bad hands, extra fingers, deformed hands” and keep the shot waist-up or close-up. If the wizard offers a “hands hidden / hands out of frame” option, use it for beginners.

They can add noise. Keep prompts plain text for best reliability, especially on mobile.

Mobile browsers sometimes block third-party cookies, aggressively clear storage, or pause tabs in the background. Any of those can invalidate your session.

“portrait, long_hair, windblown_hair” + “rim_light” for separation. If hair gets messy, remove wind and keep only long_hair + soft light.

Wide lenses imply distortion—great for dramatic perspectives but brutal for close-up portraits. If you use wide angle with a face close to camera, SDXL may stretch features and hands. Fix: either move “camera farther back” in positives (“full body, camera at distance, 24mm”) or switch to 50/85 for portraits. If it keeps distorting, add negatives like “wide angle distortion, distorted face, warped perspective”.

First, refresh the page and confirm you’re logged into the correct account. If you’re on a strict browser mode, it may block third-party content or mixed content. If it still doesn’t show, open a ticket and tell us your device + browser version.