How to beginner · 3 min read

Modal images explained

Q: Modal images explained

Modal images are managed using the modal Python package, which allows you to run GPU-accelerated image generation or processing tasks in serverless containers. You define a @app.function with GPU support and invoke it with image prompts or data to generate or manipulate images efficiently.

Quick answer

Modal images are managed using the modal Python package, which allows you to run GPU-accelerated image generation or processing tasks in serverless containers. You define a @app.function with GPU support and invoke it with image prompts or data to generate or manipulate images efficiently.

PREREQUISITES

Python 3.8+
Modal account and CLI installed
pip install modal
Basic knowledge of Python async functions

Setup

Install the modal package and configure your environment with your Modal account. This enables running GPU-backed image generation functions in the cloud.

bash

pip install modal

output

Collecting modal
  Downloading modal-1.x.x-py3-none-any.whl (xx kB)
Installing collected packages: modal
Successfully installed modal-1.x.x

Step by step

Define a Modal app with a GPU-enabled function that runs an image generation model (e.g., Stable Diffusion). Call this function with a prompt to generate an image and save it locally.

python

import modal

app = modal.App("image-gen-app")

@app.function(gpu="A10G", image=modal.Image.debian_slim().pip_install("torch", "diffusers", "transformers"))
def generate_image(prompt: str) -> str:
    from diffusers import StableDiffusionPipeline
    import torch

    pipe = StableDiffusionPipeline.from_pretrained(
        "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16
    )
    pipe = pipe.to("cuda")

    image = pipe(prompt).images[0]
    output_path = "/tmp/output.png"
    image.save(output_path)
    return output_path

if __name__ == "__main__":
    with app.run():
        path = generate_image.call("A futuristic cityscape at sunset")
        print(f"Image saved to: {path}")

output

Image saved to: /tmp/output.png

Common variations

Use different GPU types by changing gpu="A10G" to other supported GPUs.
Run image generation asynchronously with async def and await calls.
Use other image models by installing their dependencies and adjusting the pipeline.

python

import modal

app = modal.App("async-image-gen")

@app.function(gpu="A10G", image=modal.Image.debian_slim().pip_install("torch", "diffusers", "transformers"))
async def generate_image_async(prompt: str) -> str:
    from diffusers import StableDiffusionPipeline
    import torch

    pipe = StableDiffusionPipeline.from_pretrained(
        "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16
    )
    pipe = pipe.to("cuda")

    image = pipe(prompt).images[0]
    output_path = "/tmp/async_output.png"
    image.save(output_path)
    return output_path

if __name__ == "__main__":
    with app.run():
        import asyncio
        path = asyncio.run(generate_image_async.call("A serene mountain lake"))
        print(f"Async image saved to: {path}")

output

Async image saved to: /tmp/async_output.png

Troubleshooting

If you see CUDA out of memory, reduce batch size or use a smaller GPU.
Ensure your Modal CLI is logged in with modal login before running.
Check that all dependencies are installed in the modal.Image definition.

✅

Key Takeaways

Use modal.App and @app.function(gpu=...) to run GPU-backed image tasks serverlessly.
Install necessary ML libraries inside the Modal image with pip_install for smooth execution.
Invoke Modal functions synchronously or asynchronously depending on your application needs.
Troubleshoot GPU memory errors by adjusting GPU type or dependencies in your Modal function.
Modal abstracts infrastructure, letting you focus on image generation code in Python.

Verified 2026-04 · runwayml/stable-diffusion-v1-5

Verify ↗