Prompt: Transform the input image into a 3x4 grid of 12 versions of it, as if 12 different people redrew it from memory in a visual game of telephone. Clean, plain white background; even grid; each drawing centered in its cell.

The premise (the whole point): each panel is the previous one redrawn slightly worse from memory. Reading left to right, top to bottom — panel 1 (top-left) is fairly close to the original; each following panel gets more inaccurate, awkward, simplified and funny; the last few are crude, silly doodles only vaguely connected to the source.

Make every panel different — vary the art skill, line quality and the specific mistakes (some neat, some messy, some childish near-stick-figure, some wildly over-simplified or mis-remembered) — but keep them all recognizably descended from the same original. The progression from "decent" to "what is that" should read clearly across the grid and be genuinely funny.

Small panel numbers 1-12 are optional.

Avoid: every panel looking the same quality or style (they MUST visibly degrade across the grid); a panel that is just a clean copy of the original (only panel 1 is close); panels with nothing to do with the source; photoreal rendering; messy overlapping or misaligned cells; watermarks. Model: GPT Image 2 v1