Migrate from Modal

If you have been running AI models on Modal, this guide maps Modal concepts to their fal equivalents and shows how to convert your code. The core patterns are similar: both platforms use Python classes with lifecycle hooks and GPU configuration. The main differences are that fal uses HTTP endpoints instead of RPC, and fal manages container builds from a requirements list or Dockerfile rather than chaining image methods. For a broader overview of deploying existing Docker containers on fal (regardless of where they came from), see Deploy an Existing Server. If you are comparing fal to other platforms, see Migrate from Replicate or Migrate from RunPod.

Concept Mapping

Modal	fal	Notes
`@app.function()`	`@fal.function()`	Standalone serverless functions
`@app.cls()`	`class MyApp(fal.App)`	Class-based apps (recommended)
`@modal.method()`	`@fal.endpoint("/")`	fal uses HTTP endpoints, not RPC
`@modal.enter()`	`def setup(self)`	Container startup hook
`@modal.exit()`	`def teardown(self)`	Container shutdown hook
`modal.Image.debian_slim().pip_install(...)`	`requirements = [...]`	Or use `ContainerImage` for Dockerfiles
`modal.Image.from_dockerfile(...)`	`ContainerImage.from_dockerfile(...)`	Custom container support
`gpu="A100"`	`machine_type = "GPU-A100"`	GPU selection
`modal.Volume`	`/data` persistent storage	Mounted automatically on all runners
`modal.Secret`	`fal secrets set`	Secrets exposed as env vars
`modal deploy`	`fal deploy`	CLI deployment
`Cls().method.remote(x=123)`	`fal_client.subscribe(...)`	fal uses HTTP + queue, not RPC

Migration Path 1: Standalone Functions

If you use @app.function() in Modal, the closest fal equivalent is @fal.function().

Modal
fal

import modal

app = modal.App()
image = modal.Image.debian_slim().pip_install("torch", "transformers")

@app.function(image=image, gpu="A100")
def generate(prompt: str):
    from transformers import pipeline
    pipe = pipeline("text-generation", model="gpt2", device="cuda")
    return pipe(prompt)[0]["generated_text"]

@app.local_entrypoint()
def main():
    result = generate.remote(prompt="Hello world")
    print(result)

import fal

@fal.function(
    requirements=["torch", "transformers"],
    machine_type="GPU-A100",
)
def generate(prompt: str):
    from transformers import pipeline
    pipe = pipeline("text-generation", model="gpt2", device="cuda")
    return pipe(prompt)[0]["generated_text"]

Migration Path 2: Class-Based Apps (Recommended)

If you use @app.cls() with @modal.enter() and @modal.method(), convert to a fal.App class.

Modal
fal

import modal

app = modal.App()
image = (
    modal.Image.debian_slim()
    .pip_install("torch", "diffusers", "transformers", "accelerate")
)

@app.cls(image=image, gpu="A100")
class TextToImage:
    @modal.enter()
    def load_model(self):
        import torch
        from diffusers import StableDiffusionXLPipeline

        self.pipe = StableDiffusionXLPipeline.from_pretrained(
            "stabilityai/stable-diffusion-xl-base-1.0",
            torch_dtype=torch.float16,
        ).to("cuda")

    @modal.method()
    def generate(self, prompt: str):
        image = self.pipe(prompt).images[0]
        return image

    @modal.exit()
    def cleanup(self):
        del self.pipe

@app.local_entrypoint()
def main():
    result = TextToImage().generate.remote(prompt="a sunset")

import fal
from fal.toolkit import Image

class TextToImage(fal.App):
    machine_type = "GPU-A100"
    requirements = ["torch", "diffusers", "transformers", "accelerate"]

    def setup(self):
        import torch
        from diffusers import StableDiffusionXLPipeline

        self.pipe = StableDiffusionXLPipeline.from_pretrained(
            "stabilityai/stable-diffusion-xl-base-1.0",
            torch_dtype=torch.float16,
        ).to("cuda")

    @fal.endpoint("/")
    def generate(self, prompt: str) -> dict:
        image = self.pipe(prompt).images[0]
        return {"image": Image.from_pil(image)}

    def teardown(self):
        del self.pipe

Deploying

# Modal
modal deploy my_app.py

# fal
fal deploy my_app.py::TextToImage

On fal you can also validate the exact deployment locally before you ship it: fal run my_app.py::TextToImage boots the app on a temporary worker (running setup() and your endpoints just like production), so import and model-loading errors surface before they become a production crashloop.

Calling Your App

# Modal
TextToImage().generate.remote(prompt="a sunset")

# fal
import fal_client
result = fal_client.subscribe("your-username/text-to-image", arguments={
    "prompt": "a sunset"
})

Key Differences

HTTP Endpoints vs RPC

Modal uses .remote() to invoke functions as Python-to-Python RPC calls. fal uses HTTP endpoints that any language or tool can call. Your fal App exposes a REST API automatically. This means:

No .remote() calls - clients use fal_client.subscribe() or raw HTTP
Your endpoints accept and return JSON (use Pydantic models for validation)
The same endpoint works from Python, JavaScript, cURL, or any HTTP client

Queue-Based by Default

fal provides a persistent queue that handles retries, scaling, and load balancing automatically. When callers use subscribe or submit, requests go through this queue by default. Direct calls via run bypass the queue for lower overhead.

Container Images

Modal chains image methods (Image.debian_slim().pip_install(...).apt_install(...)). fal offers two approaches:

requirements list (simpler): Just list your pip packages
ContainerImage (full control): Bring your own Dockerfile

# Modal
image = (
    modal.Image.debian_slim()
    .apt_install("ffmpeg")
    .pip_install("torch", "diffusers")
)

# fal (simple)
class MyApp(fal.App):
    requirements = ["torch", "diffusers"]

# fal (Dockerfile)
from fal.container import ContainerImage

class MyApp(fal.App):
    image = ContainerImage.from_dockerfile_str("""
        FROM python:3.11
        RUN apt-get update && apt-get install -y ffmpeg
        RUN pip install torch diffusers
    """)

Volumes and Storage

Modal uses named Volume objects mounted at specific paths. fal provides /data — a persistent filesystem automatically mounted on every runner, shared across all your apps.

# Modal
volume = modal.Volume.from_name("model-cache")

@app.cls(volumes={"/cache": volume})
class MyModel:
    ...

# fal -- /data is always available, no configuration needed
class MyModel(fal.App):
    def setup(self):
        model_path = "/data/models/my-model.pt"
        if os.path.exists(model_path):
            self.model = torch.load(model_path)

Secrets

Modal uses modal.Secret.from_name(...) and attaches secrets to functions. fal exposes secrets as environment variables:

# Modal
modal secret create my-secret HF_TOKEN=hf_xxx

# fal
fal secrets set HF_TOKEN=hf_xxx

Both make secrets available via os.environ["HF_TOKEN"] in your code.

Next Steps

Once you have migrated your app, the App Lifecycle page explains how the full lifecycle works on fal, from code serialization to runner shutdown. For scaling configuration, see Scale Your Application. For monitoring your deployed app, see App Analytics.

Setting Up

Model APIs

Serverless

Compute

Organizations

Migrate from Modal

Concept Mapping

Migration Path 1: Standalone Functions

Migration Path 2: Class-Based Apps (Recommended)

Deploying

Calling Your App

Key Differences

HTTP Endpoints vs RPC

Queue-Based by Default

Container Images

Volumes and Storage

Secrets

Next Steps

​Concept Mapping

​Migration Path 1: Standalone Functions

​Migration Path 2: Class-Based Apps (Recommended)

​Deploying

​Calling Your App

​Key Differences

​HTTP Endpoints vs RPC

​Queue-Based by Default

​Container Images

​Volumes and Storage

​Secrets

​Next Steps

Concept Mapping

Migration Path 1: Standalone Functions

Migration Path 2: Class-Based Apps (Recommended)

Deploying

Calling Your App

Key Differences

HTTP Endpoints vs RPC

Queue-Based by Default

Container Images

Volumes and Storage

Secrets

Next Steps