---
title: "Image Generation"
---

> Documentation Index
> Fetch the complete documentation index at: https://truememory-docs.mvstg.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Image Generation

Truememory provides an OpenAI-compatible image generation endpoint that routes to multiple providers. Any client that works with OpenAI's `/v1/images/generations` works with Truememory by changing the base URL.

:::note[Model ID format on this endpoint]
Image models use bare OpenAI IDs or `google/`-prefixed IDs (`google/gemini-3.1-flash-image`) — not the `vertexai/` prefix chat uses. See [Model ID formats](/ai-gateway/overview/#model-id-formats).
:::


All requests require a Bearer token from [Truememory Auth](/authentication/overview/).

## Endpoint

```
POST /v1/images/generations
```

## Supported Models

| Model | Provider | Format |
|-------|----------|--------|
| `openai/gpt-image-1.5` | OpenAI | `url` or `b64_json` |
| `openai/dall-e-3` | OpenAI | `url` or `b64_json` |
| `google/gemini-3.1-flash-image` | Google Gemini ("Nano Banana 2", Vertex AI) | `b64_json` only |
| `google/gemini-3.1-flash-lite-image` | Google Gemini ("Nano Banana 2 Lite", Vertex AI) | `b64_json` only |
| `google/gemini-2.5-flash-image` | Google Gemini ("Nano Banana", Vertex AI) | `b64_json` only |
| `google/gemini-3-pro-image` | Google Gemini ("Nano Banana Pro", Vertex AI) | `b64_json` only |

Use the `provider/model` format (e.g. `openai/gpt-image-1.5`). You can also omit the provider prefix for OpenAI models (e.g. `dall-e-3`).

The Gemini image models ("Nano Banana" and "Nano Banana Pro") generate a single image per request (`n` must be `1`).

## Parameters

Send as JSON:

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `model` | string | required | Image model to use (e.g. `openai/gpt-image-1.5`, `google/gemini-3.1-flash-image`) |
| `prompt` | string | required | Text description of the desired image |
| `n` | integer | 1 | Number of images to generate. OpenAI: 1–10. Google: 1 only — Gemini image models return a single image. |
| `size` | string | null | Image size: `1024x1024`, `1792x1024`, `1024x1792`. For Google models, this maps to aspect ratios (1:1, 16:9, 9:16). |
| `quality` | string | null | Image quality (OpenAI only): `auto`, `low`, `medium`, `high`, `hd` |
| `response_format` | string | `b64_json` | `url` or `b64_json`. Google models always return `b64_json` regardless of this setting. |
| `style` | string | null | Image style (dall-e-3 only): `natural`, `vivid` |

## Examples

### OpenAI (gpt-image-1.5)

```bash
curl -X POST https://your-tie-host/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TOKEN" \
  -d '{
    "model": "openai/gpt-image-1.5",
    "prompt": "A serene mountain landscape at sunset with a reflective lake",
    "n": 1,
    "size": "1024x1024",
    "quality": "auto"
  }'
```

### Google Gemini ("Nano Banana")

```bash
curl -X POST https://your-tie-host/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TOKEN" \
  -d '{
    "model": "google/gemini-3.1-flash-image",
    "prompt": "A serene mountain landscape at sunset with a reflective lake",
    "size": "1024x1024"
  }'
```

Use `google/gemini-3-pro-image` for the higher-quality "Nano Banana Pro" model.

### DALL-E 3

```bash
curl -X POST https://your-tie-host/v1/images/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TOKEN" \
  -d '{
    "model": "openai/dall-e-3",
    "prompt": "A serene mountain landscape at sunset with a reflective lake",
    "style": "natural"
  }'
```

## Response

The response follows OpenAI's image generation format:

```json
{
  "created": 1711900000,
  "data": [
    {
      "b64_json": "iVBORw0KGgo...",
      "url": null,
      "revised_prompt": "A tranquil mountain landscape during golden hour..."
    }
  ]
}
```

| Field | Type | Description |
|-------|------|-------------|
| `created` | integer | Unix timestamp of when the response was generated |
| `data` | array | List of generated images |
| `data[].b64_json` | string or null | Base64-encoded image data (when `response_format` is `b64_json`) |
| `data[].url` | string or null | Temporary URL to the generated image (when `response_format` is `url`, OpenAI models only) |
| `data[].revised_prompt` | string or null | The prompt that was actually used (OpenAI models may revise your prompt for safety/quality) |

## Provider Differences

| Feature | OpenAI | Google Gemini (Nano Banana) |
|---------|--------|-----------------------------|
| Max images per request | 10 | 1 |
| `response_format: "url"` | Temporary URL returned | Not supported — always returns `b64_json` |
| `style` parameter | Supported (dall-e-3 only) | Ignored |
| `quality` parameter | Supported | Ignored |
| `revised_prompt` | Returned when prompt is modified | Not returned |
| Size handling | Passed as-is (e.g. `1024x1024`) | Mapped to aspect ratio (e.g. `1:1`) |

## SDK Integration

Since the endpoint is OpenAI-compatible, you can use the standard OpenAI SDK:

### Python

```python
from openai import OpenAI

client = OpenAI(
    base_url="https://your-tie-host/v1",
    api_key="$TOKEN",
)

response = client.images.generate(
    model="openai/gpt-image-1.5",
    prompt="A serene mountain landscape at sunset",
    n=1,
    size="1024x1024",
)

print(response.data[0].b64_json[:50])  # base64 image data
```

### TypeScript / Node.js

```typescript
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://your-tie-host/v1",
  apiKey: "$TOKEN",
});

const response = await client.images.generate({
  model: "openai/gpt-image-1.5",
  prompt: "A serene mountain landscape at sunset",
  n: 1,
  size: "1024x1024",
});

console.log(response.data[0].b64_json?.slice(0, 50));
```

## Error Handling

| Status | Code | When |
|--------|------|------|
| 400 | `VALIDATION_ERROR` | Unsupported model, `n` exceeds provider limit, invalid parameters |
| 401 | `AUTH_TOKEN_MISSING` | No Bearer token provided |
| 502 | `SERVICE_UPSTREAM_ERROR` | The upstream provider (OpenAI or Google) returned an error |
| 503 | `SERVICE_UNAVAILABLE` | Required provider credentials not configured (e.g. `OPENAI_API_KEY` or `GCP_PROJECT_ID`) |

Source: https://truememory-docs.mvstg.com/ai-gateway/image-generation/index.mdx