> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pindeck.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# AI generation

> Generate image variations with fal.ai and OpenRouter VLM

# AI generation

Pindeck uses AI to generate variations of your images. Choose between shot-type changes and style transformations.

## How it works

1. Select an image in the gallery
2. Click **Generate Variation**
3. Choose a variation type and detail
4. The system sends the image to fal.ai for generation
5. The result is saved as a new child image linked to the original

### Variation types

| Type        | Description                           | Example                       |
| ----------- | ------------------------------------- | ----------------------------- |
| `shot_type` | Changes camera angle/framing          | Close-up, wide shot, overhead |
| `style`     | Applies a visual style transformation | Noir, pastel, high contrast   |

## Configuration

AI generation requires two API keys set in the Convex dashboard:

| Variable                   | Description                                 |
| -------------------------- | ------------------------------------------- |
| `FAL_KEY`                  | fal.ai API key for image generation         |
| `OPENROUTER_API_KEY`       | OpenRouter API key for VLM analysis         |
| `OPENROUTER_VLM_MODEL`     | Optional model override for vision analysis |
| `OPENROUTER_PROVIDER_SORT` | Optional provider sorting preference        |

## Smart analysis

The `/smartAnalyzeImage` endpoint uses an OpenRouter VLM (vision-language model) to analyze your image before generation. This analysis helps select appropriate variation parameters.

See [Smart Analyze endpoint](/api-reference/endpoint/smart-analyze) in the API reference.

## Variation metadata

Generated images track their relationship to the original:

| Field             | Type   | Description                                   |
| ----------------- | ------ | --------------------------------------------- |
| `parentImageId`   | ID     | Reference to the source image                 |
| `variationCount`  | number | Number of variations generated                |
| `variationType`   | string | `shot_type` or `style`                        |
| `variationDetail` | string | Specific variation applied (e.g., "close-up") |
| `sourceType`      | string | Set to `ai` for generated images              |

## Rate limits and costs

* fal.ai charges per generation — monitor usage in your fal.ai dashboard
* OpenRouter charges per VLM analysis call
* Generation time varies (typically 5–15 seconds per image)