Short answer: Both write Chinese reliably. In three same-prompt tests, Qwen Image 3.0 and Nano Banana 2.1 spelled every Chinese and English word correctly. Pick Qwen Image 3.0 for Chinese posters and labels where you want bold, high-contrast type and a flat $0.0358 at 2K, plus 1–3 reference images for edits. Pick Nano Banana 2.1 for more styled, atmospheric scenes, 4K output, up to 10 references, and, until November 8, a lower price ($0.018 at 2K in HiAPI's Programmers' Day sale). Both are live on HiAPI: try Qwen Image 3.0 on its model page.
The differences at a glance
| Qwen Image 3.0 (Standard) | Nano Banana 2.1 | |
|---|---|---|
| Maker | Alibaba | |
| Model id | qwen-image-3.0/text-to-image, /image-to-image | gemini-nano-banana-2.1 |
| Price at 1K / 2K | $0.0358 / $0.0358 | $0.022 / $0.036 |
| Sale price (to Nov 8) | — | $0.011 / $0.018 |
| 4K | No | $0.058 ($0.029 on sale) |
| Reference images | 1–3, separate edit entry | Up to 10, same id |
| Aspect ratios | 8 (1:1 to 21:9) | 14 plus auto |
| Prompt rewriting | prompt_extend, on by default | No |
negative_prompt, seed | Yes | No |
Live HiAPI prices, checked October 18, 2026. Qwen Image 3.0 Pro is $0.0497 at 1K and $0.0993 at 2K.
Same prompt, side by side
Each test ran once per model at 2K, with Qwen's prompt_extend off so both saw the exact same words.
Test 1: a Chinese drinks poster
一张秋季饮品海报，3:4 竖版。画面中央是一杯透明玻璃杯装的桂花拿铁，奶泡上撒着金色桂花，
背景是浅米色纸张质感和几枝桂花。顶部大标题逐字写“秋日限定”，下方副标题写“桂花拿铁”，
底部小字写“10月18日—11月30日”。标题用优雅的中文宋体，层级清楚，留白充足，柔和的
自然光，商业海报质感。除以上文字外不要其他文字。
Qwen Image 3.0:
Generate this on Qwen Image 3.0 →
qwen-image-3.0/text-to-image, 3:4, 2K, $0.0358.
Nano Banana 2.1:
Generate this on Nano Banana 2.1 →
gemini-nano-banana-2.1, 3:4, 2K, $0.036 ($0.018 on sale).
What's visibly different: all three lines are correct on both. Qwen sets a large, bold title on a plain cream background, with osmanthus sprigs in the corners. Nano Banana 2.1 stages a fuller scene, with a layered latte on a wooden table, leaf shadows on the wall and a lighter title.
Test 2: a tea box label with small type
Product photograph of a square kraft-paper tea box on a slate surface,
front-facing. The front label reads, in Chinese, the large title “云雾绿茶”
and below it the smaller line “净含量 100 克 · 浙江产区”, plus the English
line “CLOUD MIST GREEN TEA” at the bottom. A few loose tea leaves and a
celadon cup beside the box, soft top light, crisp label print, clean
e-commerce composition. No other text.
Qwen Image 3.0:
Generate this on Qwen Image 3.0 →
qwen-image-3.0/text-to-image, 1:1, 2K, $0.0358.
Nano Banana 2.1:
Generate this on Nano Banana 2.1 →
gemini-nano-banana-2.1, 1:1, 2K, $0.036 ($0.018 on sale).
What's visibly different: both rendered the small line "净含量 100 克 · 浙江产区" correctly, the hardest part of the test. Qwen's label is denser and higher-contrast, with a rule between the lines. Nano Banana 2.1 leaves more white space on the label and uses a lighter title, a softer, more premium look.
Test 3: a bilingual shop sign in a photo
Street photograph of a small independent bookstore at dusk on an old lane,
wide 16:9 framing. A dark wooden signboard above the door reads exactly
“山海书店” in large Chinese characters with “MOUNTAIN & SEA BOOKS” in small
English capitals beneath. Warm light from the shop window, stacks of books
visible inside, a bicycle leaning on the wall, wet stone pavement. No other
readable signs.
Qwen Image 3.0:
Generate this on Qwen Image 3.0 →
qwen-image-3.0/text-to-image, 16:9, 2K, $0.0358.
Nano Banana 2.1:
Generate this on Nano Banana 2.1 →
gemini-nano-banana-2.1, 16:9, 2K, $0.036 ($0.018 on sale).
What's visibly different: both signs are correct in both languages. Qwen added a second bicycle the prompt didn't ask for. Nano Banana 2.1 kept one bicycle and added a stool, potted plants and pedestrians down the lane, a busier and more lived-in street.
Which to choose
- Chinese posters, packaging and labels with bold type: Qwen Image 3.0, $0.0358 at 2K, and turn
prompt_extendoff for exact copy. - Atmospheric scenes and lifestyle shots: Nano Banana 2.1 built fuller sets in every test.
- 4K or more than 3 reference images: Nano Banana 2.1.
negative_promptor a fixedseed: Qwen Image 3.0.- Lowest price until November 8: Nano Banana 2.1 at $0.018 for 2K. After the sale, 2K costs about the same on both ($0.036 vs $0.0358).
Switching between them
Both run on POST /v1/tasks. The differences are the model id and the reference-image field:
| Qwen Image 3.0 | Nano Banana 2.1 | |
|---|---|---|
| Generate | qwen-image-3.0/text-to-image | gemini-nano-banana-2.1 |
| Edit | qwen-image-3.0/image-to-image with image_urls (1–3) | Same id with image_input (up to 10) |
| Size | aspect_ratio + resolution (1K/2K) | aspect_ratio + resolution (1K/2K/4K) |
FAQ
Which one renders Chinese text better?
In these three tests, neither made a mistake, including a small net-weight line. The difference was style: Qwen's type is bolder and higher-contrast.
Should I turn off prompt_extend on Qwen?
For exact text, yes. It rewrites your prompt by default, which helps short prompts but can change wording.
Does Qwen Image 3.0 Pro do better?
Pro adds layout density and detail at $0.0497 for 1K. See the same-prompt test in Qwen Image 3.0 API pricing.
Code is in the Qwen Image 3.0 API guide. Or copy the tea-box prompt and run it on Qwen Image 3.0.
