Architectural Image Generation with Nano Banana 2.1: 5 Prompts and a Workflow

What will you learn in this guide?

You’ll learn how to write more controlled prompts for architectural concepts and interior images with Nano Banana 2.1, generate images from scratch, and request changes to existing images. The goal isn’t to produce technical output that verifies project dimensions or drawings; it’s to structure prompts for developing ideas, exploring atmosphere, and making visual studies before a presentation.

The Gemini API documentation uses “Nano Banana” as the name for Gemini’s image-generation and editing capabilities. It describes Nano Banana 2.1 as its high-efficiency flagship model for image generation and conversational editing. According to Google, the model includes improvements in image quality, text generation, multi-turn consistency, and grounding with Google Search, and works at 1K, 2K, and 4K resolutions. These are the producer’s product claims; they don’t guarantee the accuracy of design decisions or project data.

Requirements: access, model, and image rights

This guide is based on the API workflow. You’ll need Gemini API access and an API key to authorize requests; official examples show how to save an image generated with the Python client to a file. See the current Gemini API documentation for client code and setup details. No specific software version, price, or free usage allowance is assumed; the sources don’t explain pricing or access terms.

To generate a new image, you can send the model a text instruction. To edit an existing image, include an image input along with the text instruction. The official guide reminds users to have the necessary rights to use uploaded images and not to generate content that infringes on others’ rights.

Nano Banana 2.1 is presented as the recommended current option in the model list. The same list says Nano Banana 2 Lite is designed for use cases where speed and cost are the priority, but isn’t optimized for multiple reference inputs and successive edits. Nano Banana Pro is also listed for complex image tasks. The examples below are written for 2.1.

Five prompts for architectural images

Each example is a standalone starting prompt you can copy and adapt. The prompts are written in English and aim to specify elements such as the room, materials, lighting, and framing clearly. Prompt design is iterative: review the first result, describe only what you want to change, and try again.

1. Interior concept: Describe the type of space, material palette, lighting, and framing in the same instruction.

Create a photorealistic interior concept image of a quiet reading room in a small urban home. Use warm oak joinery, pale limestone flooring, a linen sofa, and a large north-facing window with soft overcast daylight. Show the full room from eye level, with a calm, restrained material palette. Keep the space uncluttered and do not add people, signage, or text.

This example defines the subject and atmosphere while also ruling out unwanted elements. If the materials or lighting aren’t right in the first result, explicitly change those details in a new prompt.

1. Exterior atmosphere: Specify the facade materials, surroundings, and image conditions separately.

Create a photorealistic architectural visualization of a compact contemporary house in a leafy residential setting. Show a simple two-storey volume with pale brick walls, dark metal window frames, and a recessed timber entrance. Use soft late-afternoon light after rain, with subtle reflections on the paving. Frame the building from the street at a three-quarter angle. No people, signs, or readable text.

This prompt describes a single building from a specific viewpoint. If the surroundings or facade feel out of balance in the result, narrow the framing or material description in the next iteration.

1. Changing materials in an existing image: Upload a reference image and specify which features should remain unchanged.

Edit the uploaded interior image by changing the floor finish to light terrazzo with small, subtle aggregate. Preserve the room layout, camera viewpoint, furniture positions, wall openings, and daylight direction. Do not change the ceiling design or add new objects.

The Gemini API image-editing example shows how to provide an image input alongside text. This instruction limits the edit to a single surface while asking the model to preserve other design decisions.

1. Color adjustment: Specify the desired color change instead of describing the entire image again.

In the uploaded architectural interior image, shift the wall color to a muted sage green. Keep the existing materials, furniture, composition, shadows, and lighting unchanged. Apply the color only to the painted wall surfaces; do not alter wood, stone, glass, or fabric.

This approach defines the scope of the edit. When reviewing the result, check whether the color has spread to other surfaces; if needed, specify more clearly which surfaces should be affected.

1. Consistency across multiple views: Explicitly say that you want the same design character to be preserved.

Create a second architectural interior view that matches the design language of the uploaded reference: warm oak, pale stone, soft natural light, and restrained contemporary detailing. Show the adjoining dining area from a new eye-level viewpoint. Maintain a similar material palette and calm atmosphere, but do not copy the original camera angle. No people, labels, or text.

The documentation mentions improvements to multi-turn consistency in Nano Banana 2.1; this doesn’t mean different images will match perfectly. When generating a new view from a reference, list the design features you want to preserve in the prompt and evaluate the result separately.

Common mistakes

• Giving vague instructions: General phrases such as “a modern and beautiful interior” don’t describe materials, lighting, or framing. Make the requested features specific.

• Packing too many goals into one prompt: If you need to change the layout, materials, and lighting, break the edits into steps. That makes it easier to assess which request affected the result.

• Not specifying what to preserve: In image editing, also state which camera angle, furniture arrangement, or openings you don’t want changed.

• Treating the output as technical project data: This guide focuses on visual ideation. The sources don’t say that a generated image verifies dimensions, code compliance, or construction accuracy.

• Ignoring image rights: Check that you have the right to use the reference image you upload. The Gemini API documentation also says that generated images contain a SynthID watermark.

Where it fits in an architecture and visualization workflow

Nano Banana 2.1 can help interior designers and architectural visualization teams explore alternative atmospheres or material directions. Students can also try different visual descriptions while preparing a concept presentation. However, the sources don’t explain how the model works with CAD/BIM data or whether it produces technical drawings. Treat its output as part of visual research and communication, not as a substitute for project geometry.

Before adding it to an office workflow, verify API access, pricing, and image usage terms separately; the source text doesn’t clarify these. Also check your office policy on sharing client or project images. Although high-resolution options are available, the sources don’t specify workstation requirements, so it isn’t possible to claim that any particular hardware is required.

Next steps

Start with a short prompt for a single room or facade; describe the subject, materials, lighting, and framing separately. After the first image, iterate by changing just one variable at a time. If you use a reference image, clearly distinguish the features to preserve from the area to change. Have the design team review results before using them in a presentation, and verify that you have the necessary usage rights.

Sources and license

This guide has been adapted into Turkish based on information in Google’s Gemini API documentation on image generation and prompt design. Both sources are provided under the CC BY 4.0 license: “Nano Banana image generation” and “Prompt design strategies”. The prompt examples were created specifically for this article.