sebastiankamph
patreon
Use Same Face - Ip Adapter FaceID Plus V2
🕑 Added 2024-02-08 21:53:26 +0000 UTCIP-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 
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 
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.
- 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.
- Increasing the ControlNet Control Weight (a setting of up to 1.5 can produce good results).
- Modifying the Starting and Ending Control Step.
- Changing the SD model. Some models may be biased against the face you’re trying to generate.
Comments
E
heads up- images don't work in this post
Reinhard Niedermeier
thank you, I reloaded my gallery under 'Multi Imputs' made shure to be under 'text2img' .. after that it was fine.
Reinhard Niedermeier
thank you, I cleared my images in the gallery of 'Multi-Inputs', loaded them again, and proofed to be under 'text2img' .. after that it was fine.
Sebastian Kamph
SD has a lot of errors that can be hard to troubleshoot sometimes. Make sure you've updated to the latest versions and that all input settings are correct. If the problem persists, I would try a clean install and see if you get the same error. If not, something broke, just move on to the fresh install.
Reinhard Niedermeier
thank you for your clear Infos! but what can I do, when I see this: 'AttributeError' objekt has no attribute 'mode'. ?
ckorm
Any idea how to get the multi inputs tab to work in Forge? I tried installing the main extension opposed to the built-in one but it overwrites it whenever I reload.
Coconut Pete
it could be out of date info also... I ended up keeping them .bin and they seem to be working. I wish this part of SD would be more intuitive lol
Sebastian Kamph
Haven't seen anything about this. Maybe some extensions can't handle .bins and are hardcoded for .pth? Beats me tbh. I'd test before I start renaming.
Coconut Pete
Have read some comments on Reddit and elsewhere that says I need to rename .bin files to .pth anyone know if this is correct?
Sebastian Kamph
Happy to hear you sorted it :)
ClusterFox
nvm, I think I figured it out, I had to leave control steps at 0 and 1, like the default
ClusterFox
Hi, so I use forge and while it's generating picture with the face at first, on the second phase it just becomes a totally different face, it's like in your video, but opposite, it's not pasting the face I want onto the random one, but the other way around, the random on the face I want.
Sebastian Kamph
Happy to hear you found the solution for you, and thank you for mentioning it for anyone coming after reading this!
Suaid Ishak
I'm using AWS Ubuntu EC2 instance. Was struggling to get it running in the beginning. Found out that i need in to install insightface module from extension via URL
Martin Swartling
will do...though i think i solved it
Sebastian Kamph
Thank you! Hmm, not sure how you mean. Could you post in the #help channel on discord and show screenshots?
Sebastian Kamph
Happy to hear, thanks for being a part of the journey :)
Martin Swartling
I appreciate your work royally. Having trouble getting the images to pop under this part though. Tried looking through to see if I missed something but everything seems to be in place following your instructions...do you have any suggestions on how to go about it?
Funzoomi
Man. I really love your detailed guides. Awesome work!
