Keep the model context, then adjust subject, style, lighting, camera angle, and output ratio before remixing.
Replace vague nouns with concrete subjects, locations, colors, materials, and mood words.
Run one strong version first, then iterate only the parts you want to change.
Prompt
Generate image, print it out, then crazily add handwritten Chinese annotations, doodles, scribbles with red ink. If you want, search this account's content, the doodle content mainly roasts him, can also add some small stickers
