Use Same Face - Ip Adapter FaceID Plus V2

IP-Adapter-Face-ID-Plus-V2 Automatic1111 Installation and Usage Guide

INSTALLATION

Ensure that your ControlNet extension is up to date enough that it already has the Multi-Inputs feature:

If not, go to the extensions tab:

Click on “Check for updates” and ensure that the checkbox beside “sd-webui-controlnet” before pressing “Apply and restart UI”. Wait for the extension to install completely before continuing. For best results, start a new instance of A1111.

DOWNLOADING MODELS

Go to https://huggingface.co/h94/IP-Adapter-FaceID/tree/main and download the following files by pressing the or button beside the files:

After downloading, place the .bin files in the models\ControlNet\ folder within your A1111 installation folder, and the lora.safetensors files in the models\Lora\ folder:

Using Face ID Plus V2 for the first time with an SD1.5 model

On the Automatic1111 txt2img tab, load your preferred SD1.5 model, write your prompt and modify the generation settings as usual.

Recommended settings for 1.5:

Go down to the ControlNet tab and click on the Multi-Input tab as shown here:

Click Upload Images and add several sample files of the desired face you want to apply onto your generated images, then click on the enable checkbox. The ControlNet tab should then look something like this:

Now scroll below for more ControlNet settings. For Control Type, pick IP-Adapter:

Pick ip-adapter_face_id_plus as your preprocessor and ip-adapter-faceid-plusv2_sd15 as your model:

Try the the following settings for Control Weight, Starting Control Step and Ending Control Step:

These settings allow Face ID enough control over the generation process to replace the resulting face without compromising the visual style or quality. Increase ending control step to 1 if face resemblance is lost.

With all your settings in place you can now click Generate to create your image or tweak other settings as you see fit! Face ID can run alongside other ControlNets or other Automatic1111 features such as Hires fix.

Using Face ID Plus V2 with an SDXL or SDXL Turbo model

Prompt and modify settings as normal. ControlNet model should be changed to ip-adapter-faceid-plusv2_sdxl. If you can’t see the model from the list, press the refresh button beside the dropdown menu.

Press generate button to create your desired image.

Additional Notes:

- CFG for SDXL and SDXL Turbo models is set lower than SD1.5. A setting of 1.5 to 2.0 is optimal.

- SDXL Turbo works well with very few sampling steps:

- Poor results from using Face ID (i.e. the face you’re generating doesn’t look like your input images) can be caused by many factors, experiment with the following to tweak your results.

1. Pushing the generation in the right direction by adding descriptors that apply to the input images, e.g. race, gender, distinctive features, hair color etc.
2. Increasing the ControlNet Control Weight (a setting of up to 1.5 can produce good results).
3. Modifying the Starting and Ending Control Step.
4. Changing the SD model. Some models may be biased against the face you’re trying to generate.