Troubleshooting & Glossary

The model has a fixed attention budget. When your prompt piles on adjectives, three lighting cues, two camera angles and a mood board, the model quietly picks a few and drops the rest. It's not ignoring you; it's triaging.

You'll notice specific words never showing up in results — a colour you named, a prop you asked for, a camera angle you specified. That's the tell.

Run Refine → Distill and remove any adjective that isn't earning its place. Then turn off Manifest switches you don't need. A shorter, sharper prompt is heard in full.

Without locked segments, every render is a fresh roll. Even with the same prompt, small differences compound — face, pose, framing and lighting all shift a little each time.

Two renders of the same prompt look like siblings but not twins. If you were expecting the same shot with a small change, this is what's happening.

Lock the segments you want preserved the moment you find a composition you like. Re-roll only what's left unlocked.

Midjourney's default stylize value pushes results toward its house look. High stylize values push harder — more atmosphere, more painterliness, less faithfulness to what you actually described.

Faces look slightly painted, materials look slightly fantastical, lighting feels moodier than you asked for. Nothing's wrong; the dial is just up.

Drop stylize toward the low end of the range or turn on Style Raw in the parameter deck. Raw keeps the model much closer to what you wrote.

The opposite problem. With stylize low or Raw on and no strong Logic Gate, the model has nothing to add beyond your literal words — you get a flat description rendered exactly.

The image looks correct but lifeless. Every element is there, but nothing has been designed.

Push stylize up, or — better — set a more opinionated Logic Gate. A strong Design House, Fashion House or Artistic Influence gives the model something to design with.

Adjectives like 'realistic' or 'photorealistic' don't fix this — the model already thinks it's being realistic. The problem is almost always a weak material specification or lighting that's too even and too clean.

Surfaces look injection-moulded, highlights are too soft and uniform, and even 'realistic' in the prompt doesn't shift it.

Replace vague adjectives with named materials — brushed aluminium, worn leather, wet asphalt. Then break the lighting — directional key, actual shadows. Real materials plus real light beats any amount of 'photoreal' in text.

Credits are spent on AI work — Refine, Neural Link, the Inner Atelier, in-app rendering — and on copying a finalized prompt to the clipboard, which logs it to your history and project spine. Building prompts by hand with presets, styles, the parameter deck and Manifest switches costs nothing.

You can hit zero credits and still use most of the app. What you lose is the AI reasoning, in-app rendering, and the ability to copy a finalized prompt to the clipboard.

Explore engines and presets freely. Save credits for refinement, Atelier sessions, and copying the final prompt. Note the 50 credits on signup is a one-time grant — not daily, not recurring — so plan accordingly, or move up a tier when you're doing real work.