Each slot pairs a Nano Banana prompt (conversational) and a GPT-Image prompt (structured) that generate the photograph only. Headlines, badges and labels are added afterwards in Canva or Photoshop using the post-production overlay spec — AI text rendering is still unreliable, so we never bake copy into the render. Every prompt enforces a single purple colourway (fixing the purple→blue inconsistency the audit found) and ends with a no-text/no-logo/no-watermark negative. Targets are drawn from image-audit-en.html: Slots 7 and 8 are the P0 additions (comparison/what's-in-box closer and the firmness frame); Slot 2 is the re-cast keystone; Slot 3 merges the duplicate size frames.
Resolution: generate at 2000×2000 (use AI Studio at 2K for Nano Banana; gemini.google.com caps at 1024 and needs upscaling). Reference images: the current slot is shown for context, but per Nano Banana behaviour, run the prompts text-only (no attached reference) unless you specifically want a faithful re-render — attached references over-anchor on the old colour mix.
Production target: Two balls at ~85% frame fill, slight diagonal, contact micro-shadow, pure white, brand logo on smooth ball.
Production target: Hands holding both balls; clean editorial frame for a 'smooth=deep tissue / spiky=feet' headline added in post.
Production target: Single clean grid: both balls in Standard and Large — replaces the two duplicate size frames.
Production target: 3-panel in-use (back / plantar-fascia foot / neck), single purple colourway, documentary tone.
Production target: Single figure with target zones marked; communicates breadth (neck/shoulders/back/glutes/calves/feet/hands).
Production target: Aspirational single-person lifestyle, SAME purple colourway (fixes the purple->blue inconsistency).
Production target: Square comparison vs foam roller / tennis ball (generic silhouettes) + box contents strip.
Production target: Net-new frame — sets firmness expectation to cut returns and protect the rating.
#1F4E78 headings on white or white-on-navy bars. No ALL-CAPS runs.ai-image-slotN-nano.jpg; renders save into Images/.Iteration tip: if the spike texture renders too long/sharp, add "short rounded rubber nubs, not long spikes"; if a lifestyle frame looks like stock, add "candid documentary, imperfect, natural expression, NOT posed".