Fooocus Virtual Model Guide: FaceSwap and Portrait Setup

To create a virtual-character portrait in Fooocus, supply a reference image through Input Image → Image Prompt → Advanced → FaceSwap, then describe the new scene in text. Fooocus is an SDXL image-generation application available from lllyasviel's official GitHub repository. Reference conditioning can fail, so compare each candidate with the reference before adding it to your portrait collection. readme

What are the key facts about Fooocus virtual-model workflows?

How can Fooocus use reference images for portraits?

Fooocus Image Prompt combines image inputs with a text description. The author's documentation includes single-image and multiple-image examples, with additional structural controls available through advanced options. image-prompts

For a virtual character, this gives you a concrete starting artifact: a selected reference portrait. Keep that portrait next to the resulting images so you can compare identity instead of relying on your memory.

Define the character before building a scene collection. Write a short description of the traits that should remain stable, then separate those traits from changeable elements such as clothes, lighting, and location.

An original starting prompt could be: “Studio portrait of a fictional adult character with curly dark hair, an angular face, and a neutral expression, wearing a blue shirt.” Treat it as an experiment, not a guaranteed identity formula.

Fooocus exposes FaceSwap under the advanced Image Prompt controls. It also provides separate Weight and Stop At controls for image inputs. ui

Those controls give you variables to investigate methodically. Record which input you changed and keep the other references fixed, so you can tell whether that adjustment helped the intended image.

Inpainting supplies another step for editing an existing result. The maintainer documents detail improvement, content replacement, and outpainting as related but distinct editing operations. inpaint

For a portrait collection, reserve a repair pass for a specific problem. Decide whether you need to change an accessory, improve an eye region, or replace part of an outfit before selecting the edit.

The Fooocus setup guide covers the application's broader setup. Use this workflow when the central task is judging continuity across portraits.

What limits portrait consistency in Fooocus?

The Image Prompt documentation describes failure cases in which a reference may be ignored or the result may be unrelated.

A reference-guided workflow therefore needs output review; supplying a face is not proof of consistent identity. image-prompts

Review recognizable traits across the collection. Put frontal and angled results beside the reference, and check whether the character still looks like the same person to you under the changed composition.

The application's current README describes limited long-term support and an SDXL-focused architecture.

Follow the official installation documentation for supported models and avoid assuming that a newer model family can be loaded through the same controls. readme

The source exposes guidance settings, but it does not establish a universal reference weight that preserves every identity. Start from the selected control's defaults and evaluate adjustments on your own reference. ui

How do you create a virtual model with Fooocus FaceSwap?

Install Fooocus from its official repository. Windows users can follow the archive-and-launcher instructions; the README also documents Linux environments and launch commands. readme

Once the documented environment is installed and active, launch the application from the Fooocus directory with its published entry point:

python entry_with_update.py

The launcher handles the application's startup and model-download path described in the README. Wait for that setup to finish before treating the first generation as an image-quality test. readme

Use the following reference-driven trial:

1. Generate an initial fictional adult portrait using a concise description. Choose a result whose face is clearly visible and save it as the reference for this experiment.
2. Enable Input Image, select Image Prompt, upload the portrait, and open that panel's Advanced controls. Choose FaceSwap for the portrait input. ui
3. Write a new scene prompt while keeping the character description stable. For example, try the character seated beside a window in the same plain shirt.
4. Generate a candidate and compare the face with the saved reference. Record whether the new scene worked and whether the identity remained acceptable.
5. Change only one input control or scene element for the next trial. Keep a note of Weight and Stop At settings rather than relying on an undocumented preset.

After a satisfactory scene change, test another viewing angle. Keep the original reference as the anchor, and treat a newly generated angle as a candidate that must be reviewed before it becomes another reference.

The author documents multiple image inputs and structural controls such as PyraCanny and CPDS. Introduce an additional structure reference only when you have a clear composition problem to investigate. image-prompts

Avoid changing the face reference, checkpoint, prompt style, and composition guide in the same trial. If the character changes unexpectedly, that makes it harder to identify which choice was responsible.

For a local repair, move the chosen image into Inpaint or Outpaint, mask the relevant area, and use the documented mode appropriate to the edit. Keep the unedited image for comparison. inpaint

Assess the result in two passes. First inspect the changed region; then review the entire image for continuity. A repaired accessory is only useful if the overall portrait still fits the intended collection.

For automatic regional refinement, see the separate Fooocus 2.5.0 Enhance guide. It covers a different editing stage from creating the reference portrait itself.

How do Fooocus references, LoRAs, and ComfyUI compare?

Start with references when you want to test an existing portrait directly.

If you have a compatible trained adapter, the separate Fooocus LoRA workflow explains how to load and compare it.

What else should you know about Fooocus virtual models?

Is a virtual model a separate Fooocus checkpoint?

In this portrait workflow, a virtual model is the fictional character depicted in the images. Fooocus is the image-generation application, which loads separate checkpoints and control models. readme

Can I use a reference face in Fooocus without training a LoRA?

Fooocus exposes FaceSwap at Input Image → Image Prompt → Advanced. You can try that reference-image workflow with an uploaded portrait without creating a LoRA. readme

Will Fooocus preserve the same face in every output?

Fooocus's Image Prompt documentation describes failures in which references are ignored or outputs are unrelated. Review each portrait against the reference before accepting it into a character collection. image-prompts

How can I edit a small region of a Fooocus portrait?

Use Fooocus's Inpaint or Outpaint panel to mask the region and select the appropriate editing mode. The maintainer documents separate modes for detail improvement and content changes. inpaint

Sources

• Fooocus official README and installation instructions
• Author's Image Prompt documentation
• Official interface implementation
• Maintainer's inpainting documentation
• ComfyUI official repository

Related guides on PromptZone

• Fooocus 2026: The Complete Guide
• The Ultimate Guide to Fooocus Image Prompts
• Mastering Fooocus Inpainting