Sandbox - fal

Sandbox is your creative playground for testing and comparing the latest AI models across all popular media generation operations. Run any prompt through multiple models at once to quickly find the best fit for your use case. Whether you’re generating images from text, animating photos into videos, upscaling content, or creating music and sound effects or many other operations, Sandbox provides a unified interface to experiment with multiple state-of-the-art models simultaneously.

Try it

ptbZcVWIQ_fXGGHfv8Zez_0ea5ca41bdf143a29e21e30a53120672

Generate an Image

Text-to-image with Nano Banana 2, FLUX, Recraft, and more

IwzOGSbzp6e8N00QuLtFF_129417bb24f248298e95c3fa2b1b82fb

Generate a Video

Image-to-video with Veo 3.1, Kling, Sora 2, and more

SZffdb8layV90JxfN5fkP_d2048a658d9e4dc9973598f3542833a0

Generate Music

Text-to-music with ElevenLabs, Beatoven, MiniMax, and more

What you can do

Compare models side-by-side: Run the same prompt or input across multiple models to see which produces the best results for your use case
Test all major operations: Text-to-image, image editing, video generation, upscaling, background removal, audio generation, and more
Estimate costs before running: See the estimated cost for your selected models and quantity before submitting
Search your generations: Find previous results by text, image similarity, or dominant color
Share your creations: Generate shareable links with social media previews

Supported operations

Image

Operation	Description
Text to Image	Generate images from text prompts
Edit Image	Modify an image using a prompt and reference image
Edit Multi-Images	Apply edits using multiple reference images
Upscale	Increase image resolution with AI enhancement
Remove Background	Extract subjects from their backgrounds

Video

Operation	Description
Text to Video	Generate videos from text prompts
Image to Video	Animate a still image into a video
Start + End Frame	Generate a video that transitions between two images
Upscale	Enhance video resolution
Remove Background	Remove backgrounds from video clips
Video to Audio	Generate audio/sound effects for a video

Audio

Operation	Description
Text to Music	Generate music from a description
Text to Speech	Convert text into spoken audio
Text to Sound Effects	Create sound effects from descriptions

Key concepts

Canonical input

Different AI models, even when performing the same task, often have different input schemas. For example, one image generation model might accept prompt while another uses text or description. Sandbox solves this with canonical input — a standardized input format for each operation type. When you fill out the form in Sandbox, you’re providing canonical input. Behind the scenes, Sandbox automatically remaps your input to match each model’s expected schema. For example, when you enter a prompt in the image editing operation:

The canonical field is prompt
It may be remapped to text, edit_prompt, or instruction depending on the model
Image URL fields may be remapped to image, source_image, input_image, etc.

This means the actual API request sent to each model may have a slightly different input shape than what you see in Sandbox, but the semantic meaning is preserved.

Model sets

With many models available for each operation, selecting the right ones can be overwhelming. Model sets solve this by grouping models into curated collections:

State of the Art: The best-performing models for general use
Anime Style: Models optimized for anime and illustration aesthetics
Fast Generation: Models prioritizing speed over maximum quality
Custom Sets: Create your own collections for specific workflows

To use a model set:

Click the “explore sets” button in the prompt bar
Pick a pre-built set or create your own from the modal

Model sets are operation-specific — each operation type has its own available model sets.

Runs and requests

Understanding Sandbox’s structure:

Run: A single submission in Sandbox. When you click “Run”, you create one run.
Requests: Individual API calls to each model within a run. If you select 3 models with quantity 2x, your run contains 6 requests.

Each run:

Has a unique share ID for sharing
Records all input parameters (canonical input)
Tracks the status, cost, and duration of each request
Generates a preview image for social sharing

Estimated cost

Before running, Sandbox displays an estimated cost based on:

The models you’ve selected
The quantity (number of times to run each model)

The estimate appears in the footer of the prompt bar with color-coded badges:

Green: Under $1
Yellow: $1 -$ 2.50
Red: Over $2.50

The actual cost may vary based on output complexity (e.g., longer videos cost more).

Using Sandbox

Running a generation

Select the media type: Choose Image, Video, or Audio from the top tabs

Choose an operation: Select the specific operation (e.g., Text to Image, Image to Video)

Fill in the inputs: Enter your prompt and any required media inputs

Select models: Choose which models to run (or use a model set)

Set quantity: Optionally run each model multiple times

Click Run: Submit your generation

Enhancing prompts

Many operations support prompt enhancement — an AI-powered feature that expands your basic prompt into a more detailed, effective version:

Enter your initial prompt
Click the sparkle/enhance icon next to the prompt field
Sandbox uses an LLM (or VLM for image-based operations) to expand your prompt
The enhanced prompt replaces your original

Enhancement is especially useful for:

Text-to-image: Adds artistic style, lighting, and composition details
Image-to-video: Describes motion and camera movements
Video generation: Specifies pacing and transitions

Random examples

Not sure what to try? Click the dice icon next to the operation selector to fill the form with a curated example. Each click cycles through different examples for that operation.

Working with results

After a run completes, each generation appears in the canvas area. Viewing results

Click any result to open the full-screen media viewer
Navigate between results using arrow keys or on-screen controls
View detailed metadata including model, cost, and generation time

Badges on results

Fastest: Quickest generation in the run
Cheapest: Lowest cost in the run
Free: Generated using free credits

Using results as input

Hover over any image/video result
Click the “Use as input” button to apply it to a compatible operation
Sandbox automatically switches to an appropriate operation if needed

Copying and rerunning

The Input Card (visible when “Show Metadata” is enabled) provides quick actions:

Action	Description
Copy	Populate the input form with this run’s settings
Rerun	Immediately execute the same run again
Archive	Remove from Sandbox (still accessible in request history)
Share	Generate a shareable link
Bookmark	Bookmark the run for easy access later

You can also click individual fields in the input card to copy just that value to your current form.

Searching generations

Sandbox provides powerful search capabilities to find previous generations.

Text search

Type in the search bar to find generations by semantic similarity to your query. This searches the prompts and other text fields.

Image search

Click the image icon in the search bar
Upload an image
Find visually similar generations

Color search

Click the palette icon in the search bar
Select a color
Find generations with that dominant color

Filtering

Combine search with filters:

Operation Type: Show only specific operations (e.g., just video generations)
Team Member: In team workspaces, filter by who created the generation
Bookmarks: Show only bookmarked favorites
Similarity Threshold: Adjust how closely results must match your search

Free credits

Sandbox occasionally offers free credits for specific models:

Look for the gift icon with an orange dot in the interface
Click to see all models with remaining free credits
Free credits come from:
- Global campaigns (available to all users)
- User-specific grants

When you have free credits, the cost badge shows “Free” and generations are marked with a gift icon.

Free credits and free request coupons are only usable in Sandbox and the Playground — they cannot be used through the API or Workflows.

Share your generations with anyone:

Complete a run (all generations must finish)
Click the share icon on the input card
Copy the generated link

Shared runs include:

A preview image/video for social media cards
All generations from the run
Input parameters used

Recipients can view the shared run without a fal account. If they want to remix it, they can copy the inputs to their own Sandbox.

Keyboard shortcuts

Shortcut	Action
`←` `→`	Navigate between results in viewer
`Escape`	Close viewer

Using your own apps and workflows

You can add your own serverless apps and workflows to Sandbox, allowing you to compare your custom models alongside platform models.

Requirements

For your app or workflow to work in Sandbox, it must:

Accept the canonical input fields (or supported variations) for the operation type
Return the expected output fields for that operation’s media type

Sandbox automatically validates your app’s schema when you add it and will show warnings if there are compatibility issues.

Adding your apps

Click the apps icon in the operation selector bar (next to the operation dropdown)

Search for your app or workflow using the search field

Select apps to add — Sandbox validates each app’s schema against the operation requirements

Save your selection — Added apps appear in the model picker alongside platform models

Apps are added per-operation. If your app works for multiple operations (e.g., both text-to-image and image editing), you’ll need to add it to each operation separately.

When running your own apps in Sandbox, cost and duration metrics may not be available since these are tracked differently for private endpoints.

Canonical input schemas

Each operation type has a canonical input schema — the standardized fields that Sandbox uses. Your app can use either the canonical field names or any of the supported variations.

Image operations

Operation	Canonical Fields
Text to Image	`prompt`, `aspect_ratio`
Edit Image	`prompt`, `image_url`, `aspect_ratio`
Edit Multi-Images	`prompt`, `image_urls`, `aspect_ratio`
Upscale	`image_url`, `scale`
Remove Background	`image_url`

Video operations

Operation	Canonical Fields
Text to Video	`prompt`, `duration`, `aspect_ratio`
Image to Video	`prompt`, `image_url`, `duration`, `aspect_ratio`
Start + End Frame	`prompt`, `start_frame`, `end_frame`, `duration`, `aspect_ratio`
Upscale	`video_url`, `scale`
Remove Background	`video_url`
Video to Audio	`video_url`, `prompt`

Audio operations

Operation	Canonical Fields
Text to Music	`prompt`
Text to Speech	`prompt`
Text to Sound Effects	`prompt`

Supported field variations

If your app uses different field names, Sandbox can automatically remap them. The table below shows which field names are recognized for each canonical field:

Canonical Field	Accepted Variations
`prompt`	`text`, `description`, `edit_prompt`, `instruction`, `lyrics_prompt`, `script`, `text_prompt`
`image_url`	`image`, `source_image`, `input_image`, `control_image_url`, `reference_image_url`, `start_image_url`
`image_urls`	`images`, `source_images`, `input_images`, `control_images`, `reference_image_urls`
`video_url`	`video`, `source_video`, `input_video`
`start_frame`	`start_image_url`, `start_image`, `first_frame`, `first_frame_url`, `first_image_url`
`end_frame`	`end_image_url`, `end_image`, `last_frame`, `last_frame_url`, `tail_image_url`, `last_image_url`
`scale`	`upscaling_factor`, `upscale_factor`
`duration`	`video_length`

Field transforms

Some fields undergo value transformation in addition to name remapping:

Canonical Field	Transform Behavior
`aspect_ratio`	Values like `16:9`, `4:3`, `1:1` may be converted to `landscape_16_9`, `landscape_4_3`, `square_hd` for models using `image_size`
`duration`	Automatically matched to the closest available option if model has enumerated durations (e.g., `5s`, `10s`)
`scale`	May be converted to discrete values like `"2"` or `"4"` for models with `desired_increase` field

Required output fields

Your app must return the correct output structure for Sandbox to display results:

Output Type	Required Fields
Image	`images` (array), `image` (object), or `image_url`
Video	`video` (object) or `video_url`
Audio	`audio` (object) or `audio_url`

For image outputs, the standard structure is:

{
  "images": [
    {
      "url": "https://...",
      "width": 1024,
      "height": 1024,
      "content_type": "image/png"
    }
  ]
}

For video outputs:

{
  "video": {
    "url": "https://...",
    "content_type": "video/mp4"
  }
}

For audio outputs:

{
  "audio_url": "https://..."
}

Validation indicators

When adding apps, Sandbox shows validation status:

Icon	Meaning
✓ Green	Fully compatible — schema matches operation requirements
⚠ Yellow	Has warnings — may work but some fields require transformation
✕ Red	Incompatible — missing required fields or wrong output type
ℹ Blue	Has info — some fields will be automatically transformed

Apps with validation errors (red icon) cannot be added. You’ll need to update your app’s schema to match the operation requirements.

Tips for best results

Start with model sets: Typically, the default models automatically selected are the best performing models for the operation.
Enhance your prompts: Let AI expand basic ideas into detailed prompts with the “enhance prompt” button.
Compare with quantity 1x: Save costs by running single generations when comparing models
Use image search: Find similar past generations to avoid recreating work
Bookmark favorites: Mark your best results for easy access later
Check free credits: Look for the gift icon before running expensive models

​Try it

Generate an Image

Generate a Video

Generate Music

​What you can do

​Supported operations

​Image

​Video

​Audio

​Key concepts

​Canonical input

​Model sets

​Runs and requests

​Estimated cost

​Using Sandbox

​Running a generation

​Enhancing prompts

​Random examples

​Working with results

​Copying and rerunning

​Searching generations

​Text search

​Image search

​Color search

​Filtering

​Free credits

​Sharing runs

​Keyboard shortcuts

​Using your own apps and workflows

​Requirements

​Adding your apps

​Canonical input schemas

​Image operations

​Video operations

​Audio operations

​Supported field variations

​Field transforms

​Required output fields

​Validation indicators

​Tips for best results

Try it

What you can do

Supported operations

Image

Video

Audio

Key concepts

Canonical input

Model sets

Runs and requests

Estimated cost

Using Sandbox

Running a generation

Enhancing prompts

Random examples

Working with results

Copying and rerunning

Searching generations

Text search

Image search

Color search

Filtering

Free credits

Sharing runs

Keyboard shortcuts

Using your own apps and workflows

Requirements

Adding your apps

Canonical input schemas

Image operations

Video operations

Audio operations

Supported field variations

Field transforms

Required output fields

Validation indicators

Tips for best results