MOSS-Image vs Nano Banana, Flux 2 Pro, Qwen Image and Imagen 4: poster text test
The same two poster prompts through seven image models, with the text checked by OCR. One attempt per model, shown as it came back.
Text inside images is where models still slip, so we gave the same two prompts to seven of them: MOSS-Image (Base and SFT), Google's Nano Banana and Nano Banana Pro, Black Forest Labs' Flux 2 Pro, Alibaba's Qwen Image and Google's Imagen 4. MOSS-Image ran through the official MOSS API; the others ran through Replicate.
Read this as a snapshot, not a ranking. Each model got one attempt per prompt, with default settings and a 3:4 portrait ratio. We did not regenerate anything. We checked text with an automated OCR pass and compared it with the quoted copy, so a flagged line means "zoom in", not a final verdict. We did not judge style, composition or taste.
Test 1: a jazz poster with six lines of text
Event poster for a jazz night. Headline: 'BLUE HOUR JAZZ'.
Subline: 'LIVE QUARTET'. Details on three separate lines:
'FRI 07 NOV', '20:00', 'THE OLD MILL, 12 DOCK STREET'.
Deep navy background, warm yellow lettering, a saxophone silhouette.
No other text.
- MOSS-Image Base: All lines matched
- MOSS-Image SFT: Headline read as "BLUE HOUR JATT"
- Nano Banana: All lines matched
- Nano Banana Pro: All lines matched
- Flux 2 Pro: All lines matched
- Qwen Image: Headline read as "BLUE HOUR JAZL"
- Imagen 4: All lines matched
Test 2: a Chinese and English tea house sign
Photograph of a tea house storefront at dusk. Wooden sign with large
Chinese characters: '山间茶舍'. Directly below it in capitals:
'MOUNTAIN TEA HOUSE'. Small line on the glass door:
'DAILY 09:00–18:00'. No other text.
- MOSS-Image Base: Chinese sign 山间茶舍; English sign matched; Door hours matched
- MOSS-Image SFT: Chinese sign 山间茶舍; English sign matched; Door hours matched
- Nano Banana: Chinese sign read as 山间茶夤; English sign matched; Door hours garbled
- Nano Banana Pro: Chinese sign 山间茶舍; English sign matched; Door hours matched
- Flux 2 Pro: Chinese sign read as 山间茶会; English sign matched; Door hours partly garbled
- Qwen Image: Chinese sign 山间茶舍; English sign matched; Door hours matched
- Imagen 4: Chinese sign 山间茶舍; English sign matched; Door hours matched
What we can and cannot say
- On these two prompts, the Chinese characters were the dividing line. Both MOSS-Image variants, Nano Banana Pro, Qwen Image and Imagen 4 passed our check; Nano Banana and Flux 2 Pro did not.
- On the six-line English poster, only the MOSS-Image SFT and Qwen Image headlines were flagged.
- Two prompts and one attempt each is a very small sample. A different seed could change any row of these tables.
- Models differ in price, speed and licence, and we did not compare those. MOSS-Image is released under Apache-2.0.
To try the same prompts, open the studio and switch models. If a line comes back wrong, shorten it or give it more space and see how to write prompts for exact text. For the two MOSS-Image variants alone, see Base vs SFT.
This studio is an independent service and is not affiliated with the OpenMOSS team, Google, Black Forest Labs or Alibaba.