Nano Banana 2.1 Text Test: Create a Product Poster

When a product poster includes approved copy, a visually strong image is only useful if the words survive generation. A changed flavor name, invented claim, or missing call to action can turn a promising draft into another revision cycle.

We tested Nano Banana 2.1 with a fictional sparkling-tea launch poster containing five locked text lines. Two 2K outputs spelled every line correctly and added no extra copy. The compositions varied, however, and the model ignored one typography preference. The practical lesson is straightforward: explicit copy blocks can work well, but the result still needs design and text QA.

The brief: five approved lines and no filler copy

The fictional brand ORBIT TEA needed a vertical launch poster for a peach-and-yuzu sparkling tea. We chose a simple product setup—a matte-white can, navy crescent symbol, fruit, pale stone pedestal, and warm peach background—so text rendering remained the main test.

The approved copy was:

• ORBIT TEA
• PEACH + YUZU
• SPARKLING TEA
• ZERO ADDED SUGAR
• SHOP THE DROP

The prompt prohibited people, extra cans, third-party logos, watermarks, and any additional text. ZERO ADDED SUGAR is fictional campaign copy for this test, not a verified claim about a real product.

Google describes Gemini Nano Banana 2.1 as its latest high-efficiency image-generation and conversational-editing model, with improvements to visual quality, prompt adherence, multi-turn consistency, and text rendering. The same documentation lists 1K, 2K, and 4K output options. Google DeepMind's model card also identifies clear poster text and localized text rendering as intended uses, while warning that small text, long passages, and page-length layouts can still fail.

This is a focused poster test, not a general review or a comparison with other image models.

Our Nano Banana 2.1 settings

We ran the test in the WeShop Nano Banana generator on October 14, 2026. We inspected two outputs created with the same prompt and settings.

The interface labeled both results 4:5, while the delivered pixel ratio was close to rather than mathematically exact. Check and crop the final export for the destination placement.

Use a copy block inside a visual prompt

The prompt starts with the product and art direction, then isolates the exact copy and its hierarchy. The final sentences specify prohibited additions. This makes the request easier to audit than mixing product details and approved text throughout one paragraph.

Create a premium vertical ecommerce launch poster for a fictional sparkling tea brand. Hero product: one slim matte-white aluminum can with a simple navy crescent symbol, realistic condensation, peach slices and yuzu peel, on a pale stone pedestal. Warm peach-to-cream gradient background, soft studio shadows, refined editorial layout, photoreal product lighting, generous negative space.
Render exactly these five text lines, spelled exactly as written: ORBIT TEA; PEACH + YUZU; SPARKLING TEA; ZERO ADDED SUGAR; SHOP THE DROP.
Keep ORBIT TEA as the largest headline, PEACH + YUZU directly below it, SPARKLING TEA and ZERO ADDED SUGAR as small but clearly legible supporting lines, and SHOP THE DROP as a compact button-style callout. Do not add, duplicate, alter, or misspell any text. No people, no hands, no extra cans, no third-party logos, no watermark.

For a real campaign, replace the fictional copy with approved product facts. Avoid asking the model to invent nutritional, price, sustainability, or performance claims.

Result one: all five lines rendered correctly

OUTPUT 1 — EXACT COPY

The headline, flavor, product type, claim, and call to action all matched the approved copy. No additional words were visible. The image also followed the requested one-can composition, crescent mark, fruit, stone pedestal, and warm studio palette.

One instruction was not followed: ORBIT TEA appeared in a serif typeface even though the prompt requested clean sans-serif typography. That does not affect spelling, but it matters when a brand system requires a specific typographic character.

Result two: the words stayed exact while the layout changed

OUTPUT 2 — SAME PROMPT, DIFFERENT LAYOUT

The repeat output again rendered all five lines without visible misspellings or extra copy. It preserved the brand cue and general visual direction, but enlarged the can, changed the lighting, reduced the number of peach pieces, and placed the two supporting lines on the pedestal.

Two successful outputs do not prove universal text accuracy. They do show that this short, clearly separated copy block worked twice in this specific setup.

Turn the draft into a production-ready poster

Start with approved copy, not placeholder prose. Keep the number of lines small, use unambiguous punctuation, and tell the model that every line should appear once. Separate the copy block from the visual brief, then specify which line is the headline, which lines are supporting information, and which is the call to action.

After generation, read the image character by character. Check spelling, punctuation, duplication, omissions, and invented text. Then inspect product count, logo-like marks, ingredients, and composition. If the words are correct but the layout is wrong, reuse the image as a reference and request one targeted change at a time.

Do not rely on image generation for legal or factual approval. Verify every claim independently, and move final typography into a design tool when exact fonts, kerning, accessibility, or localization rules are non-negotiable. The output is a raster image rather than a layered layout with editable text objects.

For longer paragraphs or very small disclaimers, add the copy after generation. That recommendation matches the model card's warning that small text and long passages remain difficult, especially at lower resolution.

The practical takeaway

Nano Banana 2.1 handled a five-line product poster unusually cleanly in this test: both outputs preserved the exact words while producing credible ecommerce compositions. The tradeoff was layout variability and incomplete typography control. Treat it as a fast way to create and explore poster directions, then keep a human copy and design check before delivery.

Try the same workflow in the WeShop Nano Banana generator: lock the copy first, describe the hierarchy second, and inspect the exported pixels before the poster goes live.