How to Furnish an Empty Room with AI -- The Complete Workflow

Introduction

Furnishing an empty room with AI is harder than it looks. Most people open their favorite AI image generator, upload a photo of their blank living room, type something like "make this room nice" or "furnish this room in modern style" -- and get back something generic. The furniture sits at the wrong scale. The lighting does not match. The walls shift color unexpectedly. The output looks like a stock image, not like their actual space.

The problem is not the AI. The problem is the prompt. Modern image models are capable of producing photorealistic interior designs that look like they belong in an architecture magazine -- but only if you give them the right structure. This guide walks through a proven 5-step workflow that consistently produces great results, no matter the room, no matter the style. It uses a free copyable system prompt you can paste into Claude, ChatGPT, or Gemini, combined with Nano Banana Pro inside the Rangy desktop app for the actual image generation.

No subscription to an interior design service, no hiring a decorator to visualize concepts, no wrestling with render software. Just a photo, a few answers, and one generate button.

Why Nano Banana Pro is the Best Model for Interiors

After testing every major image model on the same set of empty rooms -- living rooms, bedrooms, studios, offices -- one consistently pulled ahead for interior design work: Nano Banana Pro from Google. It is available in Rangy alongside 11 other models, and for interiors specifically, it is the one I reach for first.

Three things make it stand out:

• Architectural preservation. When you tell it to keep the walls, windows, floor, and ceiling exactly as shown, it listens. Other models tend to subtly shift the room shape or replace windows with doors. Nano Banana Pro keeps the bones of the room intact.
• Scale accuracy. It reads furniture measurements from the prompt and places items at believable proportions. A 3.5m desk reads as 3.5m, not a toy version squeezed against a wall.
• Lighting coherence. The light direction from the original photo carries through to the generated furniture. Shadows fall the right way. Materials reflect correctly.

It is not the only good model -- Seedream 4.5, Flux Kontext Pro, and Qwen 2511 all produce solid results -- but Nano Banana Pro is the most precise and photorealistic for this specific task. If you want to compare results side by side, see our guide to the best AI image generators for desktop.

The 5-Step AI Interior Design Workflow

The workflow itself is simple once you see it laid out. Each step exists to solve a specific failure mode that derails generic prompts.

The Complete System Prompt -- Copy and Use It

This is the system prompt that powers the entire workflow. Copy it, paste it into Claude, ChatGPT, or Gemini, then upload a photo of your empty room. The model will walk you through the process automatically.

You are an interior designer and AI prompt engineer. I'll share a photo of an empty room. Follow this process exactly:

STEP 1 — Analyze the room in 5 bullets:
- Architectural style and era
- Natural light direction, warmth, time of day
- Floor, wall, ceiling materials
- Window and door placement
- Constraints (radiators, low ceiling, narrow shape, awkward angles)

STEP 2 — Ask me these 6 questions, then STOP and wait:
1. Purpose? (living, bedroom, office, dining, studio, etc.)
2. How many people use it? (solo, couple, family of 4, team of 6, etc.)
3. Style preference, OR say "suggest top 3": Scandinavian / Mid-Century / Japandi / Industrial / Boho / Modern Luxury / Warm Minimal / Traditional
4. Color palette: warm neutrals / cool neutrals / earthy / moody dark / bright & airy
5. Mood in one word: cozy / luxurious / minimal / energetic / serene
6. One must-have element (reading nook, gallery wall, plant, quote, etc.) — "don't know" is valid

STEP 3 — Write the final prompt(s) under 100 words each, following these rules:

ALWAYS include:
- Opens with: "Furnish this empty room. Keep walls, windows, floor, ceiling, and architecture exactly as shown."
- 4–6 furniture pieces (small/standard room) OR 8–12 pieces with layered details (large/tall-ceiling room 3m+)
- SCALE ANCHORS for every major piece:
  * Explicit measurements ("3.5m desk," "2.4m bookshelf")
  * Architecture references ("reaching the ceiling molding," "spanning two-thirds of the wall")
  * Proportional references ("coffee table two-thirds the sofa width")
- If quantity > 3: state the number twice AND use spatial grouping ("3 on left wall, 3 on right")
- For period styles, include era-authentic details (Persian rugs + brass lamps for pre-war NYC)
- Lighting matching Step 1's analysis
- Closes with: "Wide-angle interior photograph, photorealistic, natural lighting."

WALL RULE:
- Default: architecture and wall color preserved
- Exception: if chosen style strongly implies a different wall color (pre-war NYC = forest green/oxblood, Industrial = exposed brick, Moody = deep tones), ASK first: "This style works best with [color] walls — change or keep?"

MUST-HAVE FALLBACK:
- If user said "don't know," pick ONE cozy anchor piece fitting the style (reading chair, oversized plant, gallery wall).

If I said "suggest top 3," write 3 prompts (one per style) plus a one-line recommendation.

STEP 4 — After I test the output, ask: "What needs fixing? (scale / count / proportion / swap / mood)" and refine.

STEP 5 — EDIT PROMPTS (when I want to modify the generated image):

Edit prompts follow DIFFERENT rules than generation prompts:

RULE 1 — Never re-describe existing furniture. Modern edit models preserve unlisted regions by default. Listing what to "keep" causes attention dilution and drift.

RULE 2 — Only describe the change itself, with:
- Placement anchor ("against the left wall, between two windows")
- Size anchor ("1.2m vanity," "reaching window header height")
- Material + style matching the existing scene
- Styling details (what sits on it, what's around it)

RULE 3 — Three edit types, three patterns:
- ADDITIVE ("add a vanity"): just describe the new item + placement
- REPLACEMENT ("swap the sofa"): name only the old and new item
- GLOBAL ("change wall color to green"): state only the change

RULE 4 — Keep the closing "Wide-angle interior photograph, photorealistic, natural lighting." for style consistency.

RULE 5 — If the edit fails, don't add more preservation language. Instead: describe the new item more precisely, or strengthen the placement anchor.

The prompt is long on purpose. Every rule in it exists to prevent a specific failure mode. The next sections walk through what each step is doing and why it matters. If you want to sharpen your general prompt skills, our guide to writing better AI prompts covers the fundamentals that feed into this workflow.

Step 1: Analyze the Room with AI Vision

The first step is the one most people skip, and it is the most important. Before the AI suggests any style, any furniture, or any palette, it reads the actual room you uploaded. Five bullets: architecture and era, light direction, materials, window and door placement, and constraints.

Why does this matter? Because a Scandinavian style looks different in a low-ceiling apartment than it does in a sun-drenched loft. In a compact, north-facing room with warm oak floors, Scandinavian reads as bright wool throws and pale wood furniture pulled close together. In a 3.5m-tall loft with east-facing windows and polished concrete, Scandinavian reads as sparse, linear furniture with long shadows and intentional negative space. Same style name, completely different execution.

Skipping this analysis is why "make my room Scandinavian" usually produces something generic. The AI has no information about the room's actual constraints, so it falls back to the average Pinterest image of a Scandinavian living room -- which probably does not match your space at all.

The analysis also surfaces constraints the prompt needs to respect: a radiator under the window blocks sofa placement, a narrow shape rules out a coffee table, a low ceiling means tall bookshelves will look cramped. These constraints become scale anchors in the final prompt.

Step 2: Answer the 6 Questions

After analyzing the room, the AI asks six questions. Not twelve, not twenty -- six. Each one is load-bearing.

1. Purpose. A living room and a home office need completely different furniture even if the room itself is identical. This answer drives the entire furniture list.
2. How many people use it. This is the question most prompt guides skip, and it quietly controls everything. A solo reading room has one armchair and a lamp. A family living room has a 3-seater sofa plus two accent chairs. A 6-person coworking studio has a shared desk and six chairs. Skipping this question is why AI-furnished rooms often feel scaled wrong for the life that happens in them.
3. Style preference. You can name a style directly (Scandinavian, Mid-Century, Japandi, Industrial, Boho, Modern Luxury, Warm Minimal, Traditional) or you can say "suggest top 3" and get three different proposals. The "suggest top 3" option is completely valid and often the best choice when you do not have a strong aesthetic preference yet -- it lets you compare three generated images and decide from real results.
4. Color palette. Warm neutrals, cool neutrals, earthy, moody dark, or bright and airy. Five buckets are enough. Too many palette choices cause the AI to freeze or default to beige.
5. Mood in one word. Cozy, luxurious, minimal, energetic, serene. One word forces commitment and tells the model how dense or sparse to make the space.
6. One must-have element. A reading nook, a gallery wall, a large plant, a quote above the desk. If you genuinely do not know, "don't know" is a valid answer -- the prompt has a fallback that picks one cozy anchor piece that fits the chosen style.

Step 3: Get Your Custom Prompt

Once you answer the six questions, the AI combines everything -- the room analysis from Step 1 plus your preferences from Step 2 -- into a structured image prompt under 100 words.

The prompt always opens with the architecture lock: "Furnish this empty room. Keep walls, windows, floor, ceiling, and architecture exactly as shown." This one sentence prevents the most common failure, where the AI subtly redesigns the room itself instead of just furnishing it.

Then it lists 4 to 6 furniture pieces for a standard room, or 8 to 12 pieces with layered details for large rooms with ceilings over 3 meters. Every major piece gets a scale anchor. There are three valid types:

• Explicit measurements -- "3.5m desk," "2.4m bookshelf"
• Architecture references -- "reaching the ceiling molding," "spanning two-thirds of the wall"
• Proportional references -- "coffee table two-thirds the sofa width"

Scale anchors are what separate a prompt that works from a prompt that produces toy-sized furniture floating in an oversized room. Without them, the model picks an average furniture size that may have nothing to do with your actual space.

The prompt closes with: "Wide-angle interior photograph, photorealistic, natural lighting." This locks the output into the photography style we want and prevents the model from drifting into illustration or render territory.

Step 4: Generate the Image in Rangy

Now the actual image generation. Open Rangy on Mac or Windows. If you do not have it yet, download Rangy -- the free Core tier includes 5 daily generations, enough to fully run this workflow and iterate.

Inside the app:

1. Stay on the Core tab.
2. Select Nano Banana Pro from the model list.
3. Upload your empty room photo as the reference image. This enables image-guided generation, which is what keeps the architecture anchored to the real room.
4. Paste the full prompt the AI generated in Step 3 into the prompt field.
5. Click Generate.

Rangy's 3-column layout makes this comfortable: the left column shows your history of generations, the middle handles the prompt and settings, and the right displays the result with zoom and compare mode. If you want to run the same prompt through multiple models to compare, Rangy lets you queue generations in parallel. Seedream 4.5 and Flux Kontext Pro are strong alternatives for interiors -- sometimes a specific room benefits from a second opinion.

For a deeper walkthrough of the app itself, see the Rangy documentation. For more advanced prompt work, our guide to extracting style prompts from any image pairs nicely with this workflow.

Step 5: Edit and Refine -- The Counterintuitive Rule

Once you have a first-pass image, you will almost always want to refine something. Maybe the sofa is too small, maybe the wall color drifted, maybe you want to add a vanity against a specific wall. This is where most people get it wrong.

There are three edit patterns that cover nearly every case:

1. Additive edits -- add a new item

Just describe the new item with a placement anchor and a size anchor. Do not mention anything else in the scene.

Good: "Add a 1.2m wooden mid-century vanity against the left wall, centered between the two windows, with a round rattan mirror above reaching window-header height."

Bad: "Keep the sofa and coffee table and rug exactly as they are, and add a vanity."

2. Replacement edits -- swap one item

Name only the old item and the new item. One in, one out. Nothing else.

Good: "Replace the grey sofa with a 2.4m tan leather chesterfield."

Bad: "Replace the grey sofa with a tan leather chesterfield, but keep the coffee table, the lamp, the rug, and the plant in the corner."

3. Global edits -- change one property across the scene

State only the change. The model applies it and leaves furniture untouched.

Good: "Change the wall color to warm forest green."

Bad: "Change the wall color to warm forest green while keeping all the furniture, the windows, and the floor exactly as they are."

If an edit fails, the instinct is to add more preservation language. Do not do this -- it is the wrong direction. Instead, describe the new item more precisely, or strengthen the placement anchor. "Add a vanity" might fail; "Add a 1.2m oak vanity against the left wall, 30cm from the corner, with its top at window-sill height" is far more likely to land correctly.

Common Mistakes and How to Fix Them

Four failure modes account for almost every disappointing AI interior. Each has a specific fix.

Scale drift -- furniture too small for the room

The most common failure in large or tall-ceiling rooms. You end up with a sofa that looks designed for a dollhouse floating in a loft. The fix is scale anchors -- measurements, architecture references, or proportional references for every major piece. Without them, the model defaults to an "average" furniture size that may be half of what the room needs.

AI cannot count past 3 reliably

Ask for "6 chairs" and you often get 4, 5, or 7. AI image models are famously unreliable with quantities above 3. There are two fixes:

• Spatial grouping. Instead of "6 chairs," write "3 chairs on the left wall, 3 chairs on the right." Splitting the count into two groups of 3 keeps both within the model's reliable range.
• Count-containing objects. Replace a count-sensitive plural with a single object that implies the count. Instead of "6 desks for a team of 6," write "one 3.5m shared desk seating 4 people" with 4 chairs tucked under it. The burden shifts from desks to chairs -- and even if the chair count is off, the shared desk itself reads clearly as a 4-person workspace.

Shifting the counting burden from larger, fewer items (desks) to smaller, more forgiving items (chairs) also helps. A miscounted chair is much less visually jarring than a miscounted desk.

Rooms look empty or underfurnished

Especially common in large rooms or rooms with tall ceilings. The model places 4 pieces of furniture -- enough for a 15m² studio -- in a 40m² loft, and the result looks abandoned. The fix is to scale furniture density with room size. Small and standard rooms get 4 to 6 pieces. Large rooms (3m+ ceilings or over 25m²) need 8 to 12 pieces with layered details: rugs, side tables, floor lamps, plants, art, and textile accents.

Wall colors drift or change unexpectedly

The architecture lock in the opening sentence ("Keep walls, windows, floor, ceiling, and architecture exactly as shown") handles most cases. The exception is when the chosen style strongly implies a different wall color -- pre-war NYC apartments often need forest green or oxblood walls; industrial style often calls for exposed brick; moody styles want deep, saturated tones. In those cases, the system prompt explicitly asks first: "This style works best with [color] walls -- change or keep?" This way you choose whether to override the preservation rule, instead of being surprised by it.

Why Rangy is Perfect for This Workflow

You can technically run this workflow through any AI image generator. But the workflow benefits enormously from being able to compare models side by side, iterate quickly, and keep all your edits organized locally. That is exactly what Rangy is built for.

• Nano Banana Pro plus 11 other models. Rangy includes the full roster of top image models in 2026 -- Nano Banana Pro, Nano Banana 2, Seedream 4.5 and 5 Lite, Flux Klein and Flux Kontext Pro, Qwen 2511, Kimi 2.5, GPT Image 1.5, plus upscalers and retouchers. If Nano Banana Pro struggles with a specific room, you can try Flux Kontext Pro on the same prompt in one click.
• Compare results side by side. The 3-column UI keeps your history of generations in view. You can flip between versions, compare an edit against the previous state, and zoom into details without losing the full image.
• Save edits locally. All your generated images and their metadata are saved to your own machine -- default location ~/Pictures/Eti Image Generator/ -- as PNG files with JSON sidecars. Nothing is locked inside a web dashboard.
• Video generation too. If you want to create a walkthrough of the furnished room, Rangy has a Vid tab for text-to-video generation. Pair it with the workflow in our guide to generating AI videos from text.
• Free tier. The Core tier is free and includes 5 generations per day -- plenty to run this workflow end-to-end and refine a room with a couple of edits. Download Rangy and try it with your own empty room photo.

Tips for Different Room Types

The workflow is the same regardless of room type, but each room has its own tendencies to watch for.

Living room

Anchor around the sofa. Give the sofa an explicit length in meters (2.4m for a 3-seater, 3m+ for a sectional) and make the coffee table two-thirds the sofa width. Leave walking space of at least 60cm between sofa and coffee table. For must-have elements, a reading chair paired with a floor lamp is almost always a good fallback.

Bedroom

The bed size is the dominant scale anchor. Specify queen, king, or an explicit width. Avoid asking for 4 or more decorative pillows -- AI miscounts pillows easily; two plus a throw reads cleaner. If the room has space beyond the bed, request a specific secondary anchor like "a 1m reading chair in the corner with a floor lamp" rather than vague "plus some seating."

Home office

Shift the counting burden from furniture to chairs. "One 1.8m desk with a task chair and one armchair" reads much more reliably than "a desk, a chair, a bookshelf, a filing cabinet, and a small table." If you want a bookshelf, anchor it with architecture: "a floor-to-ceiling bookshelf against the left wall."

Kitchen

Kitchens are the trickiest room to furnish with AI because they have the most rule-based elements (plumbing, outlets, appliance sizing). Anchor the island with explicit dimensions ("2.4m island with 3 counter stools"), keep appliances generic ("built-in oven, integrated dishwasher") instead of specifying brands, and avoid requesting more than 3 stools without spatial grouping.

Conclusion

Furnishing an empty room with AI does not require expensive software or a hired designer. It requires the right workflow. Analyze the room first. Define your intent with six short answers. Generate a structured prompt with scale anchors. Run it through Nano Banana Pro in Rangy. Edit with the counterintuitive rule -- describe only the change, never the preservation.

This is a free system anyone can use. Copy the system prompt, open Rangy, and try it with any empty room in your home. The first generation will likely need one or two refinements, but by your third attempt you will have an image that looks like it came from an interior design portfolio -- grounded in your actual space, scaled correctly, lit coherently, and detailed at the right density.

The tools are ready. The hard part was always the structure.