Sandbox | fal.ai Reference

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.

​

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

​

Video

​

Audio

​

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:

1. Click the “explore sets” button in the prompt bar
2. 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

​

Enhancing prompts

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

1. Enter your initial prompt
2. Click the sparkle/enhance icon next to the prompt field
3. Sandbox uses an LLM (or VLM for image-based operations) to expand your prompt
4. 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: 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

1. Click the image icon in the search bar
2. Upload an image
3. Find visually similar generations

​

Color search

1. Click the palette icon in the search bar
2. Select a color
3. 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.

​

Sharing runs

Share your generations with anyone:

1. Complete a run (all generations must finish)
2. Click the share icon on the input card
3. 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

​

Tips for best results

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