[Python] Issues Getting Ghibli-Style Image Output with Gemini API - What Am I Missing?

I'm starting a venture in AI-powered image applications, and as a learning project, I'm developing a small app that transforms input images into Ghibli-style animation art.

To experiment with different APIs, I tried the following models under the Gemini ecosystem:

1. gemini-2.0-flash-exp-image-generation – Poor results.
2. Uploading an image directly to Gemini Advanced 2.5 Pro (via chat interface) – Also produces unsatisfactory output.
3. imagen-3.0-generate-002 – Seems like this model doesn't support image input?

In contrast, when I upload the same image to ChatGPT (via the image upload feature) and use the simple prompt:

Turn this image into Ghibli-style animation art
​

…the output is significantly better and aligns more closely with the style I’m aiming for.



Question:


Am I missing something when using the Gemini API for this kind of task?

Here’s the code I used to test gemini-2.0-flash-exp-image-generation:

from PIL import Image
from io import BytesIO
import google.generativeai as genai
from google.generativeai.types import GenerateContentConfig

client = genai.Client(api_key=API_KEY)

image = Image.open("a.jpg")

prompt = "Turn this image into Ghibli-style animation art"

response = client.models.generate_content(
model='gemini-2.0-flash-exp-image-generation',
contents=[prompt, image],
config=GenerateContentConfig(
response_modalities=['Text', 'Image']
)
)

for part in response.candidates[0].content.parts:
if part.text:
print(part.text)
elif part.inline_data:
result_image = Image.open(BytesIO(part.inline_data.data))
result_image.save('output.jpg')
result_image.show()


Any tips, corrections, or best practices would be appreciated!

Thanks in advance.

Continue reading...