Chezhian Renganathan   •   about 2 years ago

Gemini pro vision model generated text in Japanese (?) out of the blue !

Hello folks,
Wondering if anyone say this behavior. I was testing Gemini pro vision model based on the example notebook. Tried a few images of food.
The response was fine.
At one attempt with a different image, the output was generated as below. Apologize if I got the language incorrectly but its strange that the model which was generating in english changed to this w/o any changes to model API call or prompt.
横断歩道は、歩行者が道路を横断するための場所です。横断歩道のない場所では、歩行者は交差点で道路を横断しなければなりません。横断歩道がある場合は、横断歩道を渡らなければなりません。横断歩道を渡る際は、左右の安全を確認し、徐行で渡ります。また、横断歩道のない場所では、左右の安全を確認し、徐行で横断します。

snippet -
...
img = PIL.Image.open('xyx.jpg')
img
response = model.generate_content(img)
to_markdown(response.text)

Response was as above in non-English!