Tutorial: Using sd-webui-EasyPhoto for Digital Avatar Creation and Video Generation on JD Cloud

Each new OpenAI release seems to disrupt an industry; the recent GPT‑4o launch is poised to spark a major shift in the emotional‑companion space, prompting entrepreneurs and creators to explore new possibilities.

The open‑source community is buzzing, expecting Hugging Face and GitHub to quickly produce the next AI "Super App" or "Killer App," following successes like the viral Miao Duck Camera and Kimi Chat, while recent Remini filters have sparked a mini‑boom in image creation.

Previously, a 10‑yuan Miao Duck Camera let users generate digital avatars in minutes, and within four months the Stable Diffusion WebUI added similar capabilities, including text‑to‑video and outfit swapping, running smoothly on a 4090 GPU. Platforms such as AutoDL and Lanrui now offer ready‑made images, and the author tested the EasyPhoto SD plugin on JD Cloud’s new AI compute platform because of its low price.

Leveraging the GPT‑4o launch, the author added a digital avatar to the demo.

Three functions were tried: digital avatar generation (similar to Miao Duck Camera), multi‑person face swapping, and text‑to‑video, all producing impressive results.

This is the multi‑face swap effect:

This is the video generation effect:

Now the step‑by‑step usage of sd-webui-EasyPhoto is introduced.

Plugin repository: https://github.com/aigc-apps/sd-webui-EasyPhoto

To save costs, JD Cloud’s AI compute platform was chosen because it already includes Stable Diffusion WebUI; simply install the EasyPhoto plugin. When launching the instance, select the Stable Diffusion application, then open it from the operation bar.

In the Extensions tab, click “Install from URL”, paste the plugin URL, and install; after a few minutes the plugin is ready.

After installation, the EasyPhoto tab appears. In Settings, set the ControlNet unit number to 4, click “Apply settings”, then “Reload UI”.

In the EasyPhoto “Train” tab, upload 5‑20 half‑body photos of yourself and click “Start Training”. After entering a user ID, training runs for about 25 minutes.

The right‑hand parameters can be left at defaults. Trained models are saved under /data/apps/stable-diffusion-webui/outputs/easyphoto-user-id-infos , accessible via JupyterLab terminal or SSH.

Once training completes, you can use the model for creative tasks. In the “Photo Inference” tab, choose a template or upload a celebrity photo for face swapping, select your digital‑avatar ID, and click “Start Generation” (≈5 minutes) to produce a personalized image. Multiple avatar models enable multi‑face swaps for humorous pairings.

In the “Video Inference” window, select a scene, choose your avatar’s user ID, and click “Start Generation” to create a video. Although the quality differs from Sora, it suffices for many scenarios.

Adjust the Prompt for better scene description via the lower‑left “Advanced Options”. Change the video format from GIF to MP4, and set “Video Max num of frames” to (seconds × fps) to produce longer videos.

This simple EasyPhoto tutorial invites readers to try it themselves and explore more fun uses, especially with the offered 2‑hour coupon.

Scan the QR code to add JD Cloud’s operation assistant on WeChat and claim the coupon.