18 May 2026

If you've spent any time on design Twitter or interior-design Reddit in the last six months, you've probably seen the screenshots — a tired living room photo on one side, the same room reimagined in five different styles on the other. Most of those mockups are coming out of Nano Banana, Google's image model that ships under the official name Gemini 2.5 Flash Image (and now Nano Banana 2 / Gemini 3.1 Flash Image, released Feb 2026).
This guide is the long version. Not the marketing brochure, but the version a designer would actually use — what the model is good at, what it still gets wrong, exact prompts that work, a workflow from reference photo to client mockup, and where it fits next to Midjourney, Freepik and the Nano Banana Pro tier.
By the end you'll have everything you need to start staging rooms in an afternoon.
Nano Banana is the friendly name for Gemini 2.5 Flash Image, Google's native multimodal image model. The Feb 2026 successor — Nano Banana 2 — runs as Gemini 3.1 Flash Image Preview and is now the default in the Gemini app.
For beginners who are completely new to Nano Banana workflows, this beginner guide explains the setup and fundamentals in a simpler way before diving into advanced interior workflows:
Nano Banana Guide For Beginners
For interior design work, three properties matter more than anything else:
Most AI image models will draw you a "modern living room" from scratch. Nano Banana takes the photo of your actual living room and rearranges it — same windows, same proportions, new sofa, new walls.
Nano Banana 2 maintains the resemblance of up to five characters and 14 objects in a single workflow.
Ask for "a Scandinavian living room in the style of Studio Oink" or "a Japandi bedroom with a tatami platform bed" and the model knows what those references mean.
Nano Banana replaces sketchbooks and Photoshop comps, not engineering drawings.
Official Gemini 2.5 Flash Image documentation and API setup guide:
Gemini 2.5 Flash Image Documentation
If you want to use Nano Banana through OpenRouter instead of Google's direct API, this setup walkthrough explains the process:
Nano Banana In OpenRouter Guide
Most failed interior-design prompts fail for the same reason: the prompter writes a paragraph of style adjectives and forgets to tell the model what to keep.
"Keeping the room layout, window position, and ceiling height the same..."
"...replace the sofa with..."
"...a low-profile bouclé sectional in cream..."
"...in a minimalist Scandinavian aesthetic..."
"...with soft natural light from the left..."
"...photorealistic, interior magazine quality, no people."
Keep the layout. Restyle this living room in Scandinavian minimalist: white walls, light oak floor, cream bouclé sofa, single statement floor lamp, one large abstract canvas above the sofa, no clutter.
Keep the layout. Restyle in Mid-Century Modern: walnut floor, olive velvet sofa, tapered wooden legs, sunburst clock on the wall, mustard yellow accent chair, warm brass lamp.
Keep the layout. Restyle in Japandi: low platform sofa in oatmeal linen, tatami-inspired rug, single ikebana arrangement, washi paper pendant lamp, walnut and oak mix.
Keep the layout. Restyle in Industrial Loft: exposed brick on one wall, dark leather chesterfield sofa, cast iron Edison bulb pendant, steel-frame coffee table, raw concrete floor.
Keep the layout. Restyle in Maximalist: deep emerald wall, velvet jewel-tone sofa, gallery wall of 12 framed prints, brass floor lamp, patterned Persian rug.
Keep the bed position and window. Restyle this bedroom in Modern Farmhouse: shiplap accent wall, white linen bedding, jute rug, reclaimed wood nightstands, single woven pendant.
Keep the bed position. Restyle in Boutique Hotel: deep charcoal walls, brass-accented headboard in tan leather, crisp white bedding with a black throw blanket, two matching brass sconces.
Keep the bed position. Restyle in Japandi: low platform bed in white oak, cream linen bedding, single washi paper pendant, ikebana arrangement on a built-in shelf, tatami rug.
Shoot the room with your phone in landscape, with the most light available.
Upload, then write the anchor first.
Edit the output instead of rewriting the whole prompt.
Build out the room gradually.
Create alternate views for spatial context.
Add paint codes, fabric brands, and source links.
Clarify that these are concept mockups, not buildable specifications.
Best Tool: Nano Banana
Best Tool: Midjourney v7
Best Tool: Freepik
Best Tool: Nano Banana Pro
Best Tool: Nano Banana on Google AI Studio
Workaround: ask for unreadable text or closed books.
Workaround: place decorative objects in front of mirrors.
Workaround: specify “same view through the window.”
Workaround: treat outputs as visual references only.
Workaround: anchor proportions explicitly.
Workaround: stay with room-wide framing.
Free through Gemini with optional paid plans.
Approximately $0.039 per image.
Approximately $0.04 per image.
Higher fidelity at higher cost.
Yes, with daily limits in Gemini.
Yes, under Google's terms.
Better text rendering, stronger consistency, improved real-world knowledge.
Not accurately enough for production work.
Use prior generated images as references.
Nano Banana edits real photos. Midjourney creates stronger from-scratch concepts.
If you've read this far, the best next step is to actually run the prompts. Take one room photo, run prompts 1–5 from the Living Room section, and see which style speaks to you. The whole exercise takes 20 minutes and changes how you think about every project after.