AI image generator repeated mistakes

That is a classic "AI mirror" issue. Most generative models struggle with left-right orientation because they view images as a whole rather than having a fixed sense of lateral direction. When you say "left," the AI often can't distinguish if you mean the character's left or the viewer's left.

Since anchoring to landmarks hasn't worked, here are the three most effective technical workarounds to force the AI to respect your layout:

Instead of using directional words like "left" or "right," use postural or action-based cues that naturally create an asymmetrical scene.

Instead of: "Add a scar on his left eye."

Try: "The lion is tilting its head, showing a scar on the eye closest to the viewer," or "The lion is reaching out with its paw to touch the marking on its shoulder."