Platform For AI:Deploy the Stepfun Step1X-Edit model

Model overview

Step1X-Edit is an open-source image editing model from Stepfun. It combines a multimodal large language model (MLLM) with a diffusion image decoder (DiT). With 19 billion parameters, Step1X-Edit provides the following capabilities:

• Precise semantic parsing: Accurately interprets editing instructions to produce results that match user intent.

• Identity consistency: Preserves subject identity during editing and keeps core features intact.

• Region-level control: Supports precise, region-level editing for fine-grained modifications.

• Broad task coverage: Supports 11 common image editing tasks, including text replacement and style transfer.

• Strong benchmark results: Achieves top scores in semantic consistency, image quality, and overall performance on the GEdit-Bench benchmark.

For more information about Step1X-Edit, see stepfun-ai/Step1X-Edit.

Resource requirements

Deploying Step1X-Edit requires a GPU with at least 48 GB of video memory.

Procedure

1. Go to the Model Gallery page.

   1. Log on to the PAI console.

   2. In the upper-left corner of the top navigation bar, select a region.

   3. In the left-side navigation pane, click Workspaces and then click the name of the workspace you want to open.

   4. In the left-side navigation pane, choose Quick Start > Model Gallery.

2. On the Model Gallery page, search for Stepfun Step1X-Edit in the model list and click the model card.

3. In the upper-right corner, click Deploy. Configure the service name and resource settings to deploy the model to Elastic Algorithm Service (EAS).

   In the deployment configuration panel, set Deployment Method to single-node standard deployment, set Resource Type to public resource, and set Instance Count to 1. Then, click Deploy.

Call the model

After deployment, call Step1X-Edit through the web application or API.

import requests
import time
EAS_URL = "<YOUR_EAS_URL>"
EAS_TOKEN = "<YOUR_EAS_TOKEN>"
class TaskStatus:
    PENDING = "pending"
    PROCESSING = "processing"
    COMPLETED = "completed"
    FAILED = "failed"
response = requests.post(
    f"{EAS_URL}/generate",
    headers={
        "Authorization": f"{EAS_TOKEN}"
    },
    json={
        "prompt": "A spaceship orbiting Earth",
        "seed": 42,
        "neg_prompt": "low quality, blurry",
        "infer_steps": 28,
        "cfg_scale": 6,
        "size":1024,
        "image": "<The Base64 encoding of your image>"
    }
)
task_id = response.json()["task_id"]
print(f"Task ID: {task_id}")
while True:
    status_response = requests.get(
        f"{EAS_URL}/tasks/{task_id}/status",
        headers={
            "Authorization": f"{EAS_TOKEN}"
        })
    status = status_response.json()
    print(f"Current status: {status['status']}")
    if status["status"] == TaskStatus.COMPLETED:
        print("Image ready!")
        break
    elif status["status"] == TaskStatus.FAILED:
        print(f"Failed: {status['error']}")
        exit(1)
    time.sleep(5)
image_response = requests.get(
    f"{EAS_URL}/tasks/{task_id}/image",
    headers={
        "Authorization": f"{EAS_TOKEN}"
    })
with open("generated_image.jpg", "wb") as f:
    f.write(image_response.content)
print("Image downloaded successfully!")

References

• PAI SDK for Python

• QuickStart billing