Is there anyway to dowload full size image

[nvcc1701]

Is there anyway to save image like download full size in the gemini ui. I see that image.save() has lower resolution than download full size.

    for i, image in enumerate(response.images):
        await image.save(path="temp/", filename=f"image_{i}.png", verbose=True)
        print(i, image)

[HanaokaYuzu]

Fixed in v1.14.2.

Now calling GeneratedImage.save will automatically download image in full size. You can optionally pass full_size=False to save 512*512 preview version instead.

[nvcc1701]

Fixed in v1.14.2.

Now calling GeneratedImage.save will automatically download image in full size. You can optionally pass full_size=False to save 512*512 preview version instead.

It seem like download full size not work. When i test download, the url of gemini_api is
url from gemini-api
but when i save in gemini ui it is like
url from gemini.google.com
and the size is not same, gemini-api(1408x768), gemini.google.com(2816×1536)

[HanaokaYuzu]

Probably due to your generated image has max width greater than 2048px. Could you please try modifying the following line locally to self.url += "=s4096" and check if it works?

Gemini-API/src/gemini_webapi/types/image.py

Line 161 in 4aba81f

self.url += "=s2048"

[nvcc1701]

Probably due to your generated image has max width greater than 2048px. Could you please try modifying the following line locally to self.url += "=s4096" and check if it works?

Gemini-API/src/gemini_webapi/types/image.py

Line 161 in 4aba81f

self.url += "=s2048"

I have tested add =s4096 it not work, it have same result like s2048. The dowload full size in gemini.google.com have higher resolution.

[HanaokaYuzu]

Can you send me the prompt you used to generate the image?

[nvcc1701]

Can you send me the prompt you used to generate the image?

Yes.

    response = await client.generate_content(
        "Ảnh con chó Kích thước z16:9.",
        model=Model.G_2_5_FLASH,
    )

[HanaokaYuzu]

So what's behind the scenes is that Google uses a "2-step" generation for every image generation request. When you send prompts in chat, the first step is triggered and a 1080p image will be generated and returned; then when the user clicks "download", another RPC will be called to trigger the second step which generates the 2k version.

This works pretty well to provide a more smooth experience on the user side as well as save cost by splitting the expensive and time-consuming generation process into 2 parts, but from the perspective of gemini_webapi, those messy RPC calls could be challenging to implement into the API and I'm still hesitating whether to integrate them into the package. That's also the reason why deep research (#80) and video generation (#84, #109) is not implemented yet: these functions all rely on extra RFCs besides a single generation request (that's how GeminiClient.generate_content works currently).

In short, I'll leave this issue open for now, while as the user base grows up probably I'll find a time to implement these features one day.

[nvcc1701]

So what's behind the scenes is that Google uses a "2-step" generation for every image generation request. When you send prompts in chat, the first step is triggered and a 1080p image will be generated and returned; then when the user clicks "download", another RPC will be called to trigger the second step which generates the 2k version.

This works pretty well to provide a more smooth experience on the user side as well as save cost by splitting the expensive and time-consuming generation process into 2 parts, but from the perspective of gemini_webapi, those messy RPC calls could be challenging to implement into the API and I'm still hesitating whether to integrate them into the package. That's also the reason why deep research (#80) and video generation (#84, #109) is not implemented yet: these functions all rely on extra RFCs besides a single generation request (that's how GeminiClient.generate_content works currently).

In short, I'll leave this issue open for now, while as the user base grows up probably I'll find a time to implement these features one day.

It's Great feature, hope to use it soon