Difference in Output Between PDF and DOC File Upload Methods

Gemini API
, , , ,
1

Hi,

I’m trying two methods for file uploads:

  1. PDF File Upload:
  2. DOC File Upload:

Despite both files containing the same content, I’m noticing differences in the output. Specifically, the output response from the DOC file upload method lacks a lot of information and steps, while the PDF method works great.

Here’s what I’m doing for each method:

1. PDF File Upload:

Since there is no support for direct PDF file upload, I’m converting each PDF page to an image, extracting text from the PDF, and then doing the following:

files.append(genai.upload_file(img))

for page, (text, image) in enumerate(zip(texts, files)):
    instruction = f"{instruction}\nPage {page+1}\n------------\n{text}\n{image}"

chat_session = model.start_chat(
    history=[
        {"role": "user", "parts": [instruction]},
    ]
)
response = chat_session.send_message(prompt)
print(response.text)

2. DOC File Upload:

For the DOC file, I’m uploading it directly to Gemini:

files.append(genai.upload_file("test.docx"))
instruction = f"Usage document:\n{file}"
chat_session = model.start_chat(
    history=[
        {"role": "user", "parts": [instruction]},
    ]
)
response = chat_session.send_message(prompt)
print(response.text)

Can anyone help me understand why there is a difference in the output between these two methods? The content in both files is the same, but the response from the DOC file upload lacks information and steps compared to the PDF method. Any insights would be appreciated!