Amir Mushich drops a mixed-media ad prompt with 40% object scale

• Amir Mushich published a Nano Banana prompt for brand ads that reduces the brief to one iconic object, one human interaction, and a hard sizing rule: the illustrated prop must take at least 40% of the canvas height prompt thread.
• The recipe is unusually specific about composition: flat single-color background, darker same-hue blobs, pure-white brush-pen illustration, and front/back depth layering so the object crosses behind and in front of the model prompt details.
• Mushich tied that visual system to a real apparel-banner stack where a client brief and 3D inputs fed Claude for prompts, Nano Banana for generation, Topaz for upscaling, and Nano Banana again for final resize to PNG outputs apparel pipeline.
• A follow-up example shows the format working as a sellable ad mockup, with a pink Puma-style composition placing a real model on an oversized drawn sneaker and matching the prompt’s “immediate readability” goal Puma example.

Mushich’s prompt is less about style adjectives than scene logic. The model asks Nano Banana to first choose a brand-defining physical object, then choose a physically believable interaction — holding it, standing on it, leaning against it, or emerging through it — before any rendering starts prompt details. That is a useful constraint for ad work because it forces a readable concept instead of a fashionable collage.

The rest of the prompt is production art direction. It locks the canvas to 1:1, caps the palette at three main colors, requires a pure white hand-drawn object, and demands z-depth integration so parts of the object sit both behind and in front of the subject. The strongest rule is scale: the object must occupy at least 40% of frame height. Mushich also specifies manga-style impact marks, motion lines, and a soft contact shadow so the cutout feels placed rather than pasted prompt details.

The Puma-style sample in a later post is basically the prompt rendered literally: monochrome pink wardrobe and background, a giant white sneaker outline, motion accents around the foot placement, and a pose built around contact with the object rather than a generic fashion stance.

Mushich says the prompt emerged from a real apparel-banner optimization job. His described stack starts with a brief plus 3D inputs, uses Claude to write prompts, then a system layer for scenes and characters, Nano Banana for generation, Topaz for upscale, and Nano Banana again for resizing into final PNG banner assets. He says the recurring pain point was character and camera fine-tuning, which suggests the prompt is doing more than styling — it is trying to standardize shot logic across many outputs apparel pipeline.

He also framed the mixed-media package as client-grade work, saying he would normally charge $3,000 to $5,000 for this kind of output and bundled it with 60+ free assets in the release post free asset post. The caveat is in his own thread: most details of the client build are withheld under NDA, so the public takeaway is the prompt architecture and stack order, not a fully documented case study NDA caveat.