Image Generation Streaming Guide

Overview

NeuroLink supports image generation through AI models like Google Vertex AI's gemini-3-pro-image-preview and gemini-2.5-flash-image. This guide explains how image generation works in both generate() and stream() modes, including CLI usage with automatic file saving, technical architecture, and usage examples.

Table of Contents

1. Architecture Overview
2. Streaming Modes
3. Image Generation Flow
4. Usage Examples
5. Implementation Details
6. Troubleshooting

---

Architecture Overview

Key Components

┌─────────────────────────────────────────────────────────────┐
│                     NeuroLink Client                         │
│  (neurolink.generate() or neurolink.stream())               │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│                    BaseProvider.stream()                     │
│  • Detects if model is image generation model               │
│  • Routes to fake streaming for image models                │
│  • Routes to real streaming for text models                 │
└────────────────────────┬────────────────────────────────────┘
                         │
         ┌───────────────┴───────────────┐
         │                               │
         ▼                               ▼
┌──────────────────┐          ┌──────────────────────┐
│  Fake Streaming  │          │   Real Streaming     │
│  (Image Models)  │          │   (Text Models)      │
└────────┬─────────┘          └──────────────────────┘
         │
         ▼
┌─────────────────────────────────────────────────────────────┐
│           Provider.executeImageGeneration()                  │
│  • Calls AI provider API with image generation config       │
│  • Returns complete image as base64                          │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│                  Yield Image Chunk                           │
│  { type: "image", imageOutput: { base64: "..." } }          │
└─────────────────────────────────────────────────────────────┘

Image Generation Models

The following models are configured for image generation:

// Supported image generation models (Vertex AI and Google AI Studio)
const IMAGE_MODELS = [
  "gemini-2.5-flash-image", // GA - Fast image generation (Nano Banana)
  "gemini-3-pro-image-preview", // Preview - High-quality with 4K support (Nano Banana Pro)
];

Important Notes:

• Image generation is supported on Google Vertex AI and Google AI Studio providers
• The gemini-3-pro-image-preview model requires location: "global" configuration on Vertex AI
• Other models can use regional endpoints like us-east5 on Vertex AI
• Images are returned as base64-encoded PNG data

---

Streaming Modes

Real Streaming vs Fake Streaming

NeuroLink uses two different streaming approaches depending on the model capabilities:

Real Streaming (Text Models)

• Uses Vercel AI SDK's native streamText() function
• Streams tokens as they are generated by the AI model
• Provides true real-time streaming experience
• Used for: GPT-4, Claude, Gemini (text), etc.

Fake Streaming (Image Models)

• Calls generate() internally to get complete result
• Yields the result progressively to simulate streaming
• Required because image generation models don't support token-by-token streaming
• Used for: gemini-2.5-flash-image, gemini-3-pro-image-preview, etc.

Why Fake Streaming?

Image generation models produce complete images, not incremental tokens. The fake streaming approach:

1. Maintains API Consistency: Same stream() interface for all models
2. Preserves User Experience: Clients can use the same code pattern
3. Enables Progressive Enhancement: Can yield text chunks before final image
4. Supports Analytics: Tracks generation time and token usage

---

Image Generation Flow

Step-by-Step Process

1. Client calls neurolink.stream()
   ↓
2. BaseProvider.stream() detects image model
   ↓
3. Routes to executeFakeStreaming()
   ↓
4. Calls this.generate() internally
   ↓
5. Provider.executeImageGeneration() is invoked
   ↓
6. AI API generates complete image
   ↓
7. Image returned as base64 string
   ↓
8. enhanceResult() preserves imageOutput field
   ↓
9. executeFakeStreaming() yields text chunks (if any)
   ↓
10. executeFakeStreaming() yields image chunk
    { type: "image", imageOutput: { base64: "..." } }
   ↓
11. Client receives and processes image chunk

Code Flow in BaseProvider

// src/lib/core/baseProvider.ts

async stream(options: StreamOptions): Promise<StreamResult> {
  // Step 1: Detect if this is an image generation model
  const isImageModel = IMAGE_GENERATION_MODELS.some((m) =>
    this.modelName.includes(m),
  );

  // Step 2: Route to fake streaming for image models
  if (isImageModel) {
    return await this.executeFakeStreaming(options, analysisSchema);
  }

  // Step 3: Use real streaming for text models
  return await this.executeRealStreaming(options, analysisSchema);
}

private async executeFakeStreaming(
  options: StreamOptions,
  analysisSchema?: z.ZodSchema,
): Promise<StreamResult> {
  // Step 4: Call generate() to get complete result
  const result = await this.generate({
    prompt: options.prompt,
    // ... other options
  });

  // Step 5: Create async generator to yield chunks
  const stream = async function* () {
    // Yield text chunks if present
    if (result.text) {
      const words = result.text.split(" ");
      for (const word of words) {
        yield { content: word + " " };
        await new Promise((resolve) => setTimeout(resolve, 50));
      }
    }

    // Step 6: Yield image chunk if present
    if (result?.imageOutput) {
      yield {
        type: "image" as const,
        imageOutput: result.imageOutput,
      };
    }
  };

  return {
    stream: stream(),
    analytics: result.analytics,
    evaluation: result.evaluation,
  };
}

---

Usage Examples

Example 1: Basic Image Generation with generate()

import { NeuroLink } from "@juspay/neurolink";

const neurolink = new NeuroLink();

// Generate an image using Vertex AI
const result = await neurolink.generate({
  input: { text: "A serene mountain landscape at sunset" },
  provider: "vertex",
  model: "gemini-3-pro-image-preview",
});

// Access the generated image
if (result.imageOutput) {
  const base64Image = result.imageOutput.base64;
  console.log(`Image generated: ${base64Image.length} characters`);

  // Save to file
  const imageBuffer = Buffer.from(base64Image, "base64");
  fs.writeFileSync("mountain.png", imageBuffer);
  console.log("✅ Image saved to mountain.png");
}

// Result also contains descriptive text
console.log("Content:", result.content);
// Output: "Generated image using gemini-3-pro-image-preview (image/png)"

// Access analytics (if enabled)
if (result.analytics) {
  console.log(`Generation time: ${result.analytics.responseTime}ms`);
  console.log(`Tokens used: ${result.analytics.usage.total}`);
}

Example 2: Image Generation with Streaming

import { NeuroLink } from "@juspay/neurolink";

const neurolink = new NeuroLink();

// Stream image generation (uses fake streaming for image models)
const result = await neurolink.stream({
  input: { text: "A futuristic city with flying cars" },
  provider: "vertex",
  model: "gemini-2.5-flash-image",
});

// Process stream chunks
for await (const chunk of result.stream) {
  if ("content" in chunk) {
    // Text chunk (description or metadata)
    process.stdout.write(chunk.content);
  } else if (chunk.type === "image") {
    // Image chunk - yielded after text chunks complete
    console.log("\n✅ Image received!");
    const base64Image = chunk.imageOutput.base64;

    // Save the image
    const imageBuffer = Buffer.from(base64Image, "base64");
    fs.writeFileSync("futuristic-city.png", imageBuffer);

    console.log(`Image size: ${imageBuffer.length} bytes`);
    console.log(`Saved to: futuristic-city.png`);
  }
}

// Access analytics after streaming completes
if (result.analytics) {
  console.log(`\nTotal generation time: ${result.analytics.responseTime}ms`);
}

Note: Image generation uses "fake streaming" - the complete image is generated first, then yielded as a single chunk. This maintains API consistency with text streaming.

Example 3: CLI Usage

# Basic image generation (saves to default path: generated-images/image-<timestamp>.png)
npx neurolink generate "A beautiful sunset over the ocean" \
  --provider vertex \
  --model gemini-3-pro-image-preview

# Output:
# 📸 Generated image saved to: generated-images/image-2025-12-16T11-50-42-209Z.png
#    Image size: 1856.34 KB
# Generated image using gemini-3-pro-image-preview (image/png)

# Generate with custom output path
npx neurolink generate "Mountain landscape at sunset" \
  --provider vertex \
  --model gemini-2.5-flash-image \
  --imageOutput ./my-images/mountain.png

# Output:
# 📸 Generated image saved to: ./my-images/mountain.png
#    Image size: 2048.67 KB
# Generated image using gemini-2.5-flash-image (image/png)

# Generate with analytics
npx neurolink generate "Futuristic city with flying cars" \
  --provider vertex \
  --model gemini-2.5-flash-image \
  --imageOutput ./images/city.png \
  --enable-analytics

# Use different models
npx neurolink generate "Serene forest scene" \
  --provider vertex \
  --model gemini-3-pro-image-preview  # Best quality, requires 'global' location

npx neurolink generate "Quick sketch of a cat" \
  --provider vertex \
  --model gemini-2.5-flash-image  # Faster generation

CLI Options:

• --imageOutput <path>: Custom path for generated image (default: generated-images/image-<timestamp>.png)
• --provider vertex or --provider google-ai: Both Vertex AI and Google AI Studio support image generation
• --model <model-name>: Image generation model to use
• --enable-analytics: Include generation metrics

Example 4: Detecting Image Chunks in Stream

import { NeuroLink } from "@juspay/neurolink";

const neurolink = new NeuroLink();

const result = await neurolink.stream({
  input: { text: "A magical forest with glowing mushrooms" },
  provider: "vertex",
  model: "gemini-2.5-flash-image",
});

let textContent = "";
let imageData: string | null = null;

for await (const chunk of result.stream) {
  // Type guard for text chunks
  if ("content" in chunk) {
    textContent += chunk.content;
  }

  // Type guard for image chunks
  if ("type" in chunk && chunk.type === "image") {
    imageData = chunk.imageOutput.base64;
    console.log("Image chunk received!");
  }
}

console.log("Text description:", textContent);
console.log("Image available:", !!imageData);

if (imageData) {
  // Process the image
  const imageBuffer = Buffer.from(imageData, "base64");
  fs.writeFileSync("magical-forest.png", imageBuffer);
  console.log(`✅ Saved ${(imageBuffer.length / 1024).toFixed(2)} KB image`);
}

Example 5: Error Handling

import { NeuroLink } from "@juspay/neurolink";

const neurolink = new NeuroLink();

try {
  const result = await neurolink.stream({
    input: { text: "A dragon flying over mountains" },
    provider: "vertex",
    model: "gemini-3-pro-image-preview",
  });

  let imageReceived = false;

  for await (const chunk of result.stream) {
    if ("type" in chunk && chunk.type === "image") {
      imageReceived = true;
      const base64Image = chunk.imageOutput.base64;

      // Validate image data
      if (!base64Image || base64Image.length === 0) {
        throw new Error("Empty image data received");
      }

      // Validate base64 format
      // Validate base64 format (padding '=' only at end, max 2 chars)
      if (!/^[A-Za-z0-9+/]*={0,2}$/.test(base64Image)) {
        throw new Error("Invalid base64 format");
      }

      // Save image
      const imageBuffer = Buffer.from(base64Image, "base64");

      // Validate minimum size (1KB)
      if (imageBuffer.length < 1024) {
        throw new Error("Image data too small");
      }

      fs.writeFileSync("dragon.png", imageBuffer);
      console.log(
        `✅ Image saved successfully (${(imageBuffer.length / 1024).toFixed(2)} KB)`,
      );
    }
  }

  if (!imageReceived) {
    console.warn("⚠️ No image was generated");
  }
} catch (error) {
  console.error("❌ Image generation failed:", error.message);

  // Handle specific error cases
  if (error.message.includes("credentials")) {
    console.error("Please check your GOOGLE_APPLICATION_CREDENTIALS");
  } else if (error.message.includes("quota")) {
    console.error("API quota exceeded");
  } else if (error.message.includes("not found")) {
    console.error("Model not available in selected region");
  }
}

Example 6: Web Application Integration

// Express.js endpoint for image generation
import express from "express";
import { NeuroLink } from "@juspay/neurolink";

const app = express();
app.use(express.json());

app.post("/api/generate-image", async (req, res) => {
  const { prompt } = req.body;

  if (!prompt) {
    return res.status(400).json({ error: "Prompt is required" });
  }

  const neurolink = new NeuroLink();

  try {
    const result = await neurolink.stream({
      input: { text: prompt },
      provider: "vertex",
      model: "gemini-2.5-flash-image",
    });

    // Set headers for streaming response
    res.setHeader("Content-Type", "text/event-stream");
    res.setHeader("Cache-Control", "no-cache");
    res.setHeader("Connection", "keep-alive");

    for await (const chunk of result.stream) {
      if ("content" in chunk) {
        // Send text chunks as SSE
        res.write(
          `data: ${JSON.stringify({ type: "text", content: chunk.content })}\n\n`,
        );
      } else if (chunk.type === "image") {
        // Send image chunk as SSE
        res.write(
          `data: ${JSON.stringify({
            type: "image",
            base64: chunk.imageOutput.base64,
            size: chunk.imageOutput.base64.length,
          })}\n\n`,
        );
      }
    }

    res.write("data: [DONE]\n\n");
    res.end();
  } catch (error) {
    res.status(500).json({
      error: error.message,
      details: "Image generation failed",
    });
  }
});

// REST endpoint (non-streaming)
app.post("/api/generate-image-sync", async (req, res) => {
  const { prompt } = req.body;

  const neurolink = new NeuroLink();

  try {
    const result = await neurolink.generate({
      input: { text: prompt },
      provider: "vertex",
      model: "gemini-2.5-flash-image",
    });

    if (result.imageOutput) {
      res.json({
        success: true,
        base64: result.imageOutput.base64,
        content: result.content,
        size: Buffer.from(result.imageOutput.base64, "base64").length,
      });
    } else {
      res.status(500).json({ error: "No image generated" });
    }
  } catch (error) {
    res.status(500).json({ error: error.message });
  }
});

app.listen(3000, () => {
  console.log("Server running on http://localhost:3000");
});

---

Implementation Details

Provider-Specific Implementation

Vertex AI provider implements image generation through the REST API:

// src/lib/providers/googleVertex.ts

async executeImageGeneration(
  options: TextGenerationOptions,
): Promise<EnhancedGenerateResult> {
  const { GoogleAuth } = await import("google-auth-library");

  // Authenticate with Google Cloud
  const auth = new GoogleAuth({
    scopes: ["https://www.googleapis.com/auth/cloud-platform"],
  });

  const client = await auth.getClient();
  const accessToken = await client.getAccessToken();

  // Determine location based on model
  const location = this.modelName.includes("gemini-3-pro-image")
    ? "global"  // gemini-3-pro-image-preview requires global
    : this.location;  // Other models can use regional endpoints

  // Build request with response modalities for image generation
  const requestBody = {
    contents: [{
      role: "user",
      parts: [{ text: options.prompt }],
    }],
    generation_config: {
      response_modalities: ["TEXT", "IMAGE"],  // CRITICAL for image generation
      temperature: options.temperature || 0.7,
      candidate_count: 1,
    },
  };

  // Call Vertex AI API
  const url = `https://${location}-aiplatform.googleapis.com/v1/projects/${this.projectId}/locations/${location}/publishers/google/models/${this.modelName}:generateContent`;

  const response = await fetch(url, {
    method: "POST",
    headers: {
      Authorization: `Bearer ${accessToken.token}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify(requestBody),
  });

  const data = await response.json();

  // Extract image from response
  const candidate = data.candidates?.[0];
  const imagePart = candidate?.content?.parts?.find(
    (part) =>
      (part.inlineData || part.inline_data) &&
      ((part.inlineData?.mimeType || part.inline_data?.mime_type)?.startsWith("image/"))
  );

  if (!imagePart) {
    throw new Error("No image generated in response");
  }

  // Extract base64 data (handle both camelCase and snake_case)
  const imageData = imagePart.inlineData?.data || imagePart.inline_data?.data;
  const mimeType = imagePart.inlineData?.mimeType || imagePart.inline_data?.mime_type || "image/png";

  // Return result with imageOutput
  const result: EnhancedGenerateResult = {
    content: `Generated image using ${this.modelName} (${mimeType})`,
    imageOutput: {
      base64: imageData,
    },
    provider: this.providerName,
    model: this.modelName,
    usage: {
      input: this.estimateTokenCount(options.prompt),
      output: 0,
      total: this.estimateTokenCount(options.prompt),
    },
  };

  // Enhance with analytics/evaluation if enabled
  return await this.enhanceResult(result, options, startTime);
}

Key Implementation Details:

1. Authentication: Uses Google Cloud service account credentials
2. Location Handling: Automatically selects global for gemini-3-pro-image-preview
3. Response Modalities: Sets ["TEXT", "IMAGE"] to enable image generation
4. Base64 Extraction: Handles both inlineData and inline_data formats
5. Result Enhancement: Preserves imageOutput through analytics pipeline

Type Definitions

// src/lib/types/streamTypes.ts

export type StreamResult = {
  stream: AsyncIterable<
    | { content: string }
    | { type: "audio"; audio: AudioChunk }
    | { type: "image"; imageOutput: { base64: string } }
  >;
  // Provider information
  provider?: string;
  model?: string;
  // Usage information
  usage?: TokenUsage;
  finishReason?: string;
  // Tool integration
  toolCalls?: ToolCall[];
  toolResults?: ToolResult[];
  toolsUsed?: string[];
  // Stream metadata
  metadata?: {
    streamId?: string;
    startTime?: number;
    totalChunks?: number;
    responseTime?: number;
  };
  // Analytics and evaluation (available after stream completion)
  analytics?: AnalyticsData | Promise<AnalyticsData>;
  evaluation?: EvaluationData | Promise<EvaluationData>;
};

// src/lib/types/generateTypes.ts

export type GenerateResult = {
  content: string;
  outputs?: { text: string }; // Future extensible for multi-modal
  audio?: TTSResult;
  imageOutput?: { base64: string } | null;
  // Provider information
  provider?: string;
  model?: string;
  // Usage and performance
  usage?: TokenUsage;
  responseTime?: number;
  // Tool integration
  toolCalls?: Array<{ toolCallId: string; toolName: string; args: object }>;
  toolResults?: unknown[];
  toolsUsed?: string[];
  enhancedWithTools?: boolean;
  // Analytics and evaluation
  analytics?: AnalyticsData;
  evaluation?: EvaluationData;
};

// Note: CLI adds savedPath to imageOutput when saving images locally
// CLI-specific type (not part of core SDK):
// imageOutput?: { base64: string; savedPath?: string } | null;

// EnhancedGenerateResult extends GenerateResult with optional analytics/evaluation
export type EnhancedGenerateResult = GenerateResult & {
  analytics?: AnalyticsData;
  evaluation?: EvaluationData;
};

// CLI-specific types
export type GenerateCommandArgs = {
  input: string;
  provider?: string;
  model?: string;
  imageOutput?: string; // Custom path for generated images
  // ... other options
};

Analytics Integration

The enhanceResult() method in BaseProvider preserves the imageOutput field while adding analytics:

// src/lib/core/baseProvider.ts

protected async enhanceResult(
  result: EnhancedGenerateResult,
  options: TextGenerationOptions,
  startTime: number,
): Promise<EnhancedGenerateResult> {
  const responseTime = Date.now() - startTime;

  // CRITICAL: Store imageOutput separately to ensure preservation
  const imageOutput = result.imageOutput;

  let enhancedResult = { ...result };

  // Add analytics if enabled
  if (options.enableAnalytics) {
    try {
      const analytics = await this.createAnalytics(result, responseTime, options);
      // Preserve ALL fields including imageOutput when adding analytics
      enhancedResult = { ...enhancedResult, analytics, imageOutput };
    } catch (error) {
      logger.warn(`Analytics creation failed: ${error.message}`);
    }
  }

  // Add evaluation if enabled
  if (options.enableEvaluation) {
    try {
      const evaluation = await this.createEvaluation(result, options);
      // Preserve ALL fields including imageOutput when adding evaluation
      enhancedResult = { ...enhancedResult, evaluation, imageOutput };
    } catch (error) {
      logger.warn(`Evaluation creation failed: ${error.message}`);
    }
  }

  // CRITICAL FIX: Always restore imageOutput if it existed
  if (imageOutput) {
    enhancedResult.imageOutput = imageOutput;
  }

  return enhancedResult;
}

Key Points:

• imageOutput is explicitly preserved through analytics/evaluation pipeline
• Spread operator ensures all existing fields are maintained
• Double-check restoration at the end prevents accidental loss

---

Troubleshooting

Common Issues

1. No Image Chunk Received

Symptom: Stream completes but no image chunk is yielded.

Possible Causes:

• Model is not an image generation model
• Wrong provider (only Vertex AI supports image generation)
• API credentials are invalid or missing
• Model not available in selected region

Solution:

// Verify you're using Vertex AI provider
const result = await neurolink.generate({
  input: { text: "Generate an image of a sunset" },
  provider: "vertex", // ✅ Required
  model: "gemini-3-pro-image-preview", // ✅ Valid image model
});

// NOT these:
// provider: "openai"     // ❌ Doesn't support image generation
// provider: "anthropic"  // ❌ Doesn't support image generation
// Note: "google-ai" also supports image generation with gemini-2.5-flash-image

// Verify credentials
console.log(
  "GOOGLE_APPLICATION_CREDENTIALS:",
  process.env.GOOGLE_APPLICATION_CREDENTIALS,
);
console.log("GOOGLE_VERTEX_PROJECT:", process.env.GOOGLE_VERTEX_PROJECT);

2. Empty Base64 String

Symptom: Image chunk received but base64 field is empty.

Possible Causes:

• API returned error but didn't throw
• Response format changed
• Network issue during transmission

Solution:

for await (const chunk of result.stream) {
  if (chunk.type === "image") {
    if (!chunk.imageOutput.base64) {
      console.error("Empty image data received");
      console.error("Full chunk:", JSON.stringify(chunk, null, 2));
    } else {
      console.log(`Image data length: ${chunk.imageOutput.base64.length}`);
    }
  }
}

3. Model Not Found Error

Symptom: Error: models/gemini-3-pro-image-preview is not found for API version v1

Cause: gemini-3-pro-image-preview requires location: "global" but a regional endpoint is being used.

Solution:

// The provider automatically handles location selection:
// - gemini-3-pro-image-preview → uses "global"
// - Other models → uses configured region (e.g., "us-east5")

// Set region in environment variable
process.env.GOOGLE_VERTEX_LOCATION = "us-east5"; // For non-preview models

// Or pass in options
const result = await neurolink.generate({
  input: { text: "Generate image" },
  provider: "vertex",
  model: "gemini-2.5-flash-image", // Uses regional endpoint
  region: "us-east5",
});

4. Large Image Timeout

Symptom: Generation times out for large/complex images.

Solution:

const result = await neurolink.stream({
  input: { text: "A detailed cityscape with many buildings" },
  provider: "vertex",
  model: "gemini-2.5-flash-image",
  timeout: 60000, // Increase timeout to 60 seconds
});

5. CLI Image Not Saved

Symptom: CLI shows success but no file created.

Possible Causes:

• imageOutput option not passed to processOptions()
• Directory permissions issue
• Disk space full

Solution:

# Check default location
ls -lh generated-images/

# Use custom path with explicit directory
npx neurolink generate "test" \
  --provider vertex \
  --model gemini-2.5-flash-image \
  --imageOutput ./my-images/test.png

# Check file was created
ls -lh ./my-images/test.png

# Verify directory permissions
ls -ld generated-images/

Debug Mode

Enable debug logging to troubleshoot issues:

# Set environment variable
export DEBUG=neurolink:*

# Or use CLI flag
npx neurolink generate "test image" \
  --provider vertex \
  --model gemini-2.5-flash-image \
  --debug

# Debug output will show:
# - Provider selection
# - Model configuration
# - API request details
# - Response parsing
# - Image data extraction

Testing Image Generation

Quick test to verify image generation works:

# Test with default path
npx neurolink generate "A simple red circle" \
  --provider vertex \
  --model gemini-2.5-flash-image

# Expected output:
# 📸 Generated image saved to: generated-images/image-2025-12-16T11-50-42-209Z.png
#    Image size: 234.56 KB
# Generated image using gemini-2.5-flash-image (image/png)

# Verify file exists
ls -lh generated-images/image-*.png | tail -1

# Test with custom path
npx neurolink generate "A simple blue square" \
  --provider vertex \
  --model gemini-2.5-flash-image \
  --imageOutput ./test-output/square.png

# Expected output:
# 📸 Generated image saved to: ./test-output/square.png
#    Image size: 198.34 KB
# Generated image using gemini-2.5-flash-image (image/png)

# Verify file
file ./test-output/square.png
# Output: ./test-output/square.png: PNG image data, 1024 x 1024, 8-bit/color RGB

---

Best Practices

1. Always Check for Image Chunks

let hasImage = false;

for await (const chunk of result.stream) {
  if ("type" in chunk && chunk.type === "image") {
    hasImage = true;
    // Process image
  }
}

if (!hasImage) {
  console.warn("No image was generated");
}

2. Validate Base64 Data

if (chunk.type === "image") {
  const base64 = chunk.imageOutput.base64;

  // Validate it's valid base64 (padding '=' only at end, max 2 chars)
  if (!/^[A-Za-z0-9+/]*={0,2}$/.test(base64)) {
    throw new Error("Invalid base64 data");
  }

  // Validate minimum size (e.g., 1KB)
  if (base64.length < 1000) {
    throw new Error("Image data too small");
  }
}

3. Handle Both Text and Image

let description = "";
let imageData: string | null = null;

for await (const chunk of result.stream) {
  if ("content" in chunk) {
    description += chunk.content;
  } else if (chunk.type === "image") {
    imageData = chunk.imageOutput.base64;
  }
}

// Use both description and image
console.log("Description:", description);
if (imageData) {
  saveImage(imageData);
}

4. Use Analytics for Monitoring

const result = await neurolink.stream({
  prompt: "Generate image",
  enableAnalytics: true,
});

// Monitor generation performance
if (result.analytics) {
  console.log(`Generation time: ${result.analytics.responseTime}ms`);
  console.log(`Cost: $${result.analytics.cost}`);

  // Alert if too slow
  if (result.analytics.responseTime > 30000) {
    console.warn("Image generation took longer than 30 seconds");
  }
}

---

Conclusion

NeuroLink's image generation streaming provides a unified interface for both text and image generation. The fake streaming approach ensures consistency while maintaining the benefits of streaming APIs. By following the patterns and examples in this guide, you can effectively integrate image generation into your applications.

For more information:

• API Reference
• Provider Comparison
• Provider Status Monitoring