Support direct GCS output (storageUri) for 'gemini-2.5-flash-image' to prevent OOM in high concurrency

[LukeYu22]

SDK Version

1.21.0

Model Used

gemini-2.5-flash-image

Description

I am using the google-genai SDK to generate images using the gemini-2.5-flash-image model via the client.models.generateContent method.

Currently, the SDK returns the generated image directly as raw bytes (inline data) in the response memory. While this works for single requests, it causes significant memory pressure (OOM) in high-concurrency scenarios because the application server must hold multiple large image buffers in the heap before uploading them to storage (e.g., S3/GCS).

In the legacy Vertex AI Imagen API (e.g., imagen-3.0), we could provide a storageUri parameter to instruct the model to write the output directly to Google Cloud Storage (GCS), returning only the URI. This allowed for "zero-copy" handling on the client side.

My question is:
Is there a way to configure a GCS destination (similar to storageUri or outputConfig) in GenerateContentConfig for Gemini image generation models? If not, are there any plans to support this feature?

Current Code Implementation

Here is how I am currently calling the API. The imageByte array is the bottleneck.

public GoogleVertexAiResponse imageGenByNanoBanana(GoogleVertexAiRequest aiRequest){
    // ... validation logic ...

    GenerateContentConfig.Builder contentConfigBuilder =
            GenerateContentConfig.builder()
                    .candidateCount(1)
                    .httpOptions(HttpOptions.builder()
                            .apiVersion("v1")
                            .timeout(60000)
                            .build());

    if(!StringUtils.isEmpty(aiRequest.getAspectRatio())){
        contentConfigBuilder.imageConfig(ImageConfig.builder()
                .aspectRatio(aiRequest.getAspectRatio())
                .build());
    }

    GenerateContentConfig contentConfig = contentConfigBuilder.build();

    try {
        // The response contains the full image bytes in memory
        GenerateContentResponse response =
                client.models.generateContent(
                        aiRequest.getModelVersion(), // gemini-2.5-flash-image
                        aiRequest.getPrompt(),
                        contentConfig);

        List<Part> parts =
                response
                        .candidates()
                        .flatMap(candidates -> candidates.stream().findFirst())
                        .flatMap(Candidate::content)
                        .flatMap(Content::parts)
                        .orElse(new ArrayList<>());

        byte[] imageByte = null;
        for(Part part: parts){
            // Memory bottleneck here:
            if(part.inlineData().flatMap(Blob::data).isPresent()){
                imageByte = part.inlineData().flatMap(Blob::data).get();
                break;
            }
        }
        
        // ... upload logic ...

    } catch (Exception e){
        log.error("Google Vertex Ai Image Gen Occur An Error: {}", e.getMessage(), e);
        return handleException(e);
    }
}

Expected Behavior

I would like to be able to specify a GCS URI in the config so that the response returns a reference to the file instead of the raw bytes.

Example (Hypothetical):

GenerateContentConfig.builder()
    .imageConfig(ImageConfig.builder()
        .storageUri("gs://my-bucket/output/") // feature request
        .build())
    .build();

---

Thanks

[hemasekhar-p]

Thank you for reaching out and providing description of your use case.We have created an internal bug to track this feature request.