> ## Documentation Index
> Fetch the complete documentation index at: https://api-tools.memories.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Image Model Selection

> Choose the right image understanding model for your workload

Use this page when your request is **image-only**. Memories.ai currently documents Gemini, GPT, Nova, and Qwen image understanding endpoints.

<Info>
  This guide is intentionally practical. It focuses on the request and output patterns you will actually use in Memories.ai, not external benchmark scoreboards.
</Info>

## Quick Picks

<CardGroup cols={2}>
  <Card title="Best default" icon="brain" href="/visual-intelligence/gemini/gemini-ilm">
    Start with **Gemini Image** if you want one strong default for image reasoning and schema-based extraction.
  </Card>

  <Card title="Simplest JSON path" icon="code" href="/visual-intelligence/gpt/gpt-ilm">
    Start with **GPT Image** if you want the most familiar documented `response_format` path for JSON output.
  </Card>

  <Card title="Lowest-cost batch" icon="bolt" href="/visual-intelligence/qwen/qwen-ilm">
    Start with **Qwen Image** if cost is the main constraint and you need image analysis at scale.
  </Card>

  <Card title="Tool-first extraction" icon="wrench" href="/visual-intelligence/nova/nova-ilm">
    Start with **Nova Image** if your workflow is built around tool-style extraction via `toolConfig`.
  </Card>
</CardGroup>

## Choose By Workflow

| If your goal is...                                 | Start here                                                                                                 | Why                                                                                                 |
| -------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------- |
| General-purpose image reasoning                    | [Gemini Image](/visual-intelligence/gemini/gemini-ilm)                                                     | Strong default when you need reasoning plus schema-style structured output                          |
| Strict JSON extraction with a familiar API pattern | [GPT Image](/visual-intelligence/gpt/gpt-ilm)                                                              | The page documents `response_format` directly and uses an OpenAI-style image request/response shape |
| Lowest-cost image analysis                         | [Qwen Image](/visual-intelligence/qwen/qwen-ilm)                                                           | Cheapest published starting price among the documented image providers                              |
| Tool-driven extraction pipeline                    | [Nova Image](/visual-intelligence/nova/nova-ilm)                                                           | The docs expose `toolConfig` for structured extraction through tool specs                           |
| Explicit thinking toggle plus schema output        | [Gemini Image](/visual-intelligence/gemini/gemini-ilm) or [Qwen Image](/visual-intelligence/qwen/qwen-ilm) | Both document structured output plus explicit thinking-related fields                               |

## Provider Differences That Matter

| Provider                                               | Documented input shape                                                       | Structured output path                            | Thinking control                      | When it usually fits best                                                  |
| ------------------------------------------------------ | ---------------------------------------------------------------------------- | ------------------------------------------------- | ------------------------------------- | -------------------------------------------------------------------------- |
| [Gemini Image](/visual-intelligence/gemini/gemini-ilm) | `input_file` with image MIME type                                            | `extra_body.metadata.responseSchema`              | `thinking_config.thinking_budget`     | Default image reasoning with schema-based extraction                       |
| [GPT Image](/visual-intelligence/gpt/gpt-ilm)          | `image_url` blocks in OpenAI-style messages                                  | `response_format`                                 | Not documented on this page           | Image-only JSON extraction with the simplest request/response mental model |
| [Nova Image](/visual-intelligence/nova/nova-ilm)       | `image_url` plus text; tool config lives in `extra_body.metadata.toolConfig` | Tool-based extraction via `toolConfig`            | Not documented on this page           | Tool-oriented extraction flows and low-cost operational usage              |
| [Qwen Image](/visual-intelligence/qwen/qwen-ilm)       | `image` plus `text` blocks in the same content array                         | `extra_body.metadata.response_format.json_schema` | `enable_thinking` + `thinking_budget` | Lowest-cost image analysis with explicit thinking controls                 |

## Suggested Starting Models

| Scenario                                  | Recommended model               |
| ----------------------------------------- | ------------------------------- |
| Default image reasoning / extraction      | `gemini:gemini-3-flash-preview` |
| Familiar JSON-first image extraction      | `gpt:gpt-5-mini`                |
| Cheapest first pass on large image volume | `qwen:qwen3-vl-flash`           |
| Tool-oriented extraction workflow         | `nova:us.amazon.nova-lite-v1:0` |

<Note>
  Qwen source pricing is documented per `1K` tokens on its provider pages. This guide converts the lowest published tier into `1M` token terms for faster cross-provider comparison.
</Note>

## How To Evaluate On Your Own Data

Before standardizing on an image model, compare at least these cases:

* clean product or document images vs noisy real-world photos
* captioning vs field extraction
* free-form output vs strict JSON output
* small interactive workloads vs bulk batch processing

The right choice often depends more on your output contract than on headline model branding.

## Related Guides

* [Video Model Selection](/visual-intelligence/video-model-selection)
* [Gemini Image](/visual-intelligence/gemini/gemini-ilm)
* [GPT Image](/visual-intelligence/gpt/gpt-ilm)
* [Nova Image](/visual-intelligence/nova/nova-ilm)
* [Qwen Image](/visual-intelligence/qwen/qwen-ilm)