A few weeks ago, we wrote about one-shotting full web apps with gr.HTML: building rich, interactive frontends entirely inside Gradio using custom HTML, CSS, and JavaScript. That unlocked a lot. But what if that's not enough?
What if you want to build with your own frontend framework entirely like React, Svelte, or even plain HTML/JS, while still benefiting from Gradio's queuing system, API infrastructure, MCP support, and ZeroGPU on Spaces?
That's exactly the problem gradio.Server solves. And it changes what's possible with Gradio and Hugging Face Spaces.
Text Behind Image : an editor where you upload a photo, the background gets removed using an ML model, and then you place stylized text between the foreground subject and the background. The text appears to sit behind the person or object in the image.
This needs:
There's no way to express this UI in Gradio components. It's a full web application. But We still wanted the backend power of Gradio: queuing, concurrency management, ZeroGPU support, and hosting on HF Spaces without infrastructure headaches.
gradio.Server
gradio.Server extends FastAPI. It gives you the full power of FastAPI (custom routes, middleware, file uploads, any response type) while adding Gradio's API engine on top: queuing, SSE streaming, concurrency control, and gradio_client compatibility.
Here's the entire backend for Text Behind Image:
import os
import torch
from PIL import Image
from torchvision import transforms
from transformers import AutoModelForImageSegmentation
from gradio import Server
from gradio.data_classes import FileData
from fastapi.responses import HTMLResponse
import spaces
torch.set_float32_matmul_precision("high")
birefnet = AutoModelForImageSegmentation.from_pretrained(
"ZhengPeng7/BiRefNet", trust_remote_code=True
)
birefnet.to("cuda")
birefnet.float()
transform_image = transforms.Compose([
transforms.Resize((1024, 1024)),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
])
app = Server()
@spaces.GPU
def segment(image: Image.Image) -> Image.Image:
"""Run BiRefNet segmentation to produce a transparency mask."""
image_size = image.size
input_images = transform_image(image).unsqueeze(0).to("cuda")
with torch.no_grad():
preds = birefnet(input_images)[-1].sigmoid().cpu()
pred = preds[0].squeeze()
mask = transforms.ToPILImage()(pred).resize(image_size)
image.putalpha(mask)
return image
@app.api(name="remove_background")
def remove_background(image_path: FileData) -> FileData:
"""Remove background from an image. Returns transparent PNG."""
im = Image.open(image_path["path"]).convert("RGB")
result = segment(im)
out_path = image_path["path"].rsplit(".", 1)[0] + ".png"
result.save(out_path)
return FileData(path=out_path)
@app.get("/", response_class=HTMLResponse)
async def homepage():
html_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "index.html")
with open(html_path, "r", encoding="utf-8") as f:
return f.read()
app.launch(show_error=True)
That's it. ~50 lines of Python. The model loads at startup, @spaces.GPU handles ZeroGPU allocation, and gradio.Server manages queuing and concurrency. Let's break down what's happening.
@app.api() Instead of a Plain FastAPI Route?
If this were a regular FastAPI app, you'd define a @app.post() route for background removal. That works, until two users hit it at once. Without concurrency management, both requests fight for the GPU, and the app crashes or returns garbage.
@app.api() solves this. It wraps your function with Gradio's queuing engine: requests are serialized, concurrency is controlled, and on ZeroGPU Spaces, GPU allocation is handled automatically via @spaces.GPU. As a bonus, any @app.api() endpoint is also callable via gradio_client, so other apps or scripts can use your Space programmatically:
from gradio_client import Client, handle_file
client = Client("ysharma/text-behind-image")
result = client.predict(
image_path=handle_file("photo.jpg"),
api_name="/remove_background"
)
Meanwhile, @app.get("/") is a standard FastAPI route that serves the HTML page. Both coexist naturally because Server is a FastAPI app.
The index.html in this example is a self-contained ~1300-line web application. No React, no build step, no bundler. Just vanilla HTML with:
z-index<canvas> compositingThe frontend talks to the backend using the Gradio JS Client:
import { Client, handle_file } from "https://cdn.jsdelivr.net/npm/@gradio/client/dist/index.min.js";
const client = await Client.connect(window.location.origin);
const result = await client.predict("/remove_background", {
image_path: handle_file(file),
});
foregroundLayer.src = result.data[0].url; // transparent PNG
This is the key part: by using the Gradio JS client rather than a raw fetch() call, the frontend goes through Gradio's queue. That means concurrency is managed, GPU requests don't collide, and you could even show queue position or progress to the user. Everything else, text rendering, layer compositing, export, happens in the browser.
Here's what was not possible before gradio.Server:
| Before | After |
|---|---|
| Custom UI meant leaving Gradio entirely | Custom UI with Gradio's backend engine |
| No way to serve static HTML from a Gradio app | @app.get("/") serves anything |
gradio_client only worked with Gradio component apps |
@app.api() endpoints are client-compatible |
| Choosing between Gradio's infra and design freedom | You get both |
With gradio.Server, Gradio doubles as a backend framework, use its UI system when you want it, bring your own frontend when you don't.
If you want Gradio's UI, you can use gr.Blocks, gr.Interface, gr.ChatInterface. If you want your own UI, use gradio.Server and bring whatever frontend you like. Either way, you get Spaces hosting, API queuing, gradio_client access, the full HF ecosystem, and more.
The app is live on Spaces: ysharma/text-behind-image
Upload any photo with a clear subject, and start layering text behind it. Try the 3D extrusion, perspective tilt, and stroke effects, they combine nicely.
This post covered the core idea: gradio.Server lets you pair any frontend with Gradio's backend. There's more to explore, including MCP tool registration with @app.mcp.tool(), SSE streaming for real-time updates, batch processing, and patterns for building multi-page apps with shared state.
We'll dig into those in a follow-up post. Stay tuned.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。