Text-in-image rendering is the ability of an image generation model to produce legible, correctly spelled text within generated images — on signs, labels, posters, screens, clothing, or any surface. This has historically been one of the hardest problems in image generation: early models produced garbled letters, misspellings, or characters that resembled text but weren't readable. Modern approaches use language model reasoning to plan the text layout, font sizing, and character shapes before the image generation step, treating text as a structured constraint rather than a visual texture.
Before generating the image, Nano Banana Pro uses Gemini 3 Pro's reasoning mode to plan where text should appear, how large each character should be relative to the canvas, and how the text should wrap or curve to fit the target surface. This pre-planning step prevents common text errors like overlapping characters or text running off the edge.
Render not just single words but full sentences and short paragraphs. Menu boards, poster copy, and product descriptions with multiple lines of correctly formatted text — proper line breaks, consistent spacing, and aligned margins.
Generate text in English, Chinese, Japanese, Korean, and Latin-alphabet European languages. The model handles different scripts' character complexities — CJK characters, accented Latin letters, and mixed-language text within the same image.
Nano Banana Pro can ground text content against Google Search results — useful for generating images with real brand names, accurate product specs, or factual quotes. The model retrieves correct spellings and formats from web data.
Text renders appropriately on different surfaces: flat on signs, curved on mugs, embossed on leather, printed on fabric, displayed on screens. The model matches perspective distortion, material texture, and lighting to make text look physically present rather than overlaid.
Go to Sunra Image and select Nano Banana Pro from the model dropdown.
Include the text you want rendered in quotation marks within your prompt: *"A coffee shop chalkboard menu that reads 'Espresso $4 · Latte $5 · Cappuccino $5' in white chalk lettering."* The model renders the quoted text verbatim.
Describe where the text should appear and its visual treatment: *"Bold sans-serif white text on the storefront window"* or *"Handwritten script on the bottom of the postcard."* Be explicit about font style, color, and position.
Click Generate and zoom in on the rendered text. Check spelling, character completeness, and readability at the intended viewing size. Regenerate if any characters are malformed.
If the text is mostly correct but one word is off, use Nano Banana Pro's precision editing to fix just the text without regenerating the entire image.
Whether you're a solo creator, an agency, or a brand — every model adapts to how you work.
Fashion editorial duo
Two models stand back-to-back in a white studio. The oversized T-shirt on the left model reads "AVANT GARDE" in bold black sans-serif across the chest. The right model's tote bag has "MAISON DE MODE" in thin gold embroidered script. Clean editorial lighting, full body framing.
Tropical poolside candid
A person lounges on a rattan daybed by a turquoise pool, reading a paperback. The book cover facing the camera shows the title "THE LONG SHORE" in white serif text on a navy background. A frosted glass on the side table has "SUNSET BAR" etched into the surface. Golden hour light, shallow depth of field.
Ceramics workshop candid
A potter shapes a bowl on a spinning wheel in a sunlit workshop. The canvas apron reads "CLAY & FIRE STUDIO" in stamped block letters across the chest pocket. A chalkboard on the wall behind says "OPEN WORKSHOP — SATURDAYS 10AM" in white chalk. Warm natural light from a side window.
A rustic chalkboard mounted on a brick wall. Written in white chalk: 'DAILY SPECIALS' at the top in large letters, followed by 'Drip Coffee $3.50' and 'Cold Brew $4.50' and 'Matcha Latte $5.00'. Small chalk drawing of a coffee cup in the corner. Warm café lighting.
Model: Nano Banana Pro · Reasoning: Thinking · Resolution: 1024×1024
A minimalist book cover with a dark navy background. The title 'THE QUIET HOURS' in large gold serif letters centered on the cover. Below in smaller white text: 'A Novel by Sarah Chen'. Subtle embossed texture on the background. Clean, editorial design.
Model: Nano Banana Pro · Reasoning: Thinking · Resolution: 768×1024
A cylindrical tea canister with a matte green finish. The label reads 'SENCHA' in large English letters at the top, '煎茶' in Japanese characters below, and 'Premium Green Tea · 100g' at the bottom. Minimalist Japanese design aesthetic. White background studio shot.
Model: Nano Banana Pro · Reasoning: Thinking · Resolution: 1024×1024
A glass storefront window with gold vinyl lettering that reads 'BAKER & BLOOM' in an elegant sans-serif font. Below in smaller text: 'Artisan Bakery · Est. 2019'. Blurred interior visible through the glass. Morning sunlight reflects on the window. Street-level perspective.
Model: Nano Banana Pro · Reasoning: Thinking · Resolution: 1024×768
Generate shareable quote images with styled text — motivational quotes, book excerpts, or brand taglines rendered on designed backgrounds. No Canva template needed — describe the aesthetic and text in one prompt.
Generate product packaging with readable brand names, ingredient lists, and size labels. Useful for pitch decks and concept validation before investing in physical packaging design.
Create restaurant menu boards, store signage, and wayfinding concepts with correct text. Faster than mockup tools for early-stage design exploration — generate 10 menu board concepts in the time it takes to build 1 in Figma.
Generate the same marketing image with text in different languages — an ad banner in English, Japanese, and Spanish. The model handles each language's typographic requirements natively.
| Nano Banana Pro | GPT Image 2 | |
|---|---|---|
| Text accuracy | High accuracy — Gemini reasoning pre-plans character layout before generation | ~99% reported accuracy — slightly higher consistency on long strings and edge cases |
| Multilingual support | English, Chinese, Japanese, Korean, and European languages with proper glyph rendering | Strong English and European languages. CJK support available but less tested |
| World knowledge grounding | Google Search grounding — retrieves correct spellings for brands, places, technical terms | Strong general knowledge from training data but no live search grounding |
| Multi-turn text editing | Single-turn generation + separate editing via precision edit mode | Built-in multi-turn editing — refine text iteratively in the same conversation |
| Best for | Multilingual text, factual content needing search grounding, surface-aware placement | Long English text, iterative refinement, text-heavy compositions with many corrections |
Always quote the text you want rendered: 'Your Text Here'. Without quotes, the model may interpret the text as a prompt instruction rather than content to render. Double-check spelling before generating — the model renders what you type.
Don't leave font choice to the model if you have a preference. Describe it: "bold sans-serif", "thin serif", "handwritten script", "monospace". Without guidance, the model picks a style that fits the scene — which may not match your brand guidelines.
Short text (1–15 words) renders most reliably. As text length increases beyond ~30 words, the chance of a misspelled or malformed character increases. For paragraph-length text, generate in sections or use the image as a design comp rather than a final asset.
If your prompt includes small text (e.g., fine print, ingredient lists), generate at the highest available resolution. At 512×512, text below ~20px equivalent will be blurry regardless of the model's accuracy. Use 1024×1024 or larger for text-heavy images.
Join thousands of creators, agencies, and brands who use Sunra every day.
Start with free daily credits. No credit card required.
