Skip to playerSkip to main content
Stop wasting hours training heavy custom LoRAs just to get consistent character faces! In this tutorial, I show you how to use IP-Adapter FaceID Plus V2 inside ControlNet to instantly lock down facial consistency using just a single reference photo.

Unlike older swap tools (like ReActor or Roop) which simply paste a flat face mask onto a finished image, IP-Adapter FaceID Plus V2 injects the facial structure directly into the diffusion generation process itself. This means it natively adapts to any aesthetic style—whether you are rendering highly realistic photos, 3D renders, cartoons, or anime! Best of all, this process is completely uncensored, works faster than standard swap extensions, and is fully compatible with Automatic1111, WebUI Forge, and ComfyUI.

I guide you step-by-step through:
* Updating your local WebUI installations and ControlNet extensions.
* Downloading the correct bin models and companion LoRAs for both SD 1.5 and SDXL pipelines.
* Configuring your ControlNet settings with the InsightFace + CLIP-H pre-processors.
* Combining face reference maps with separate thematic concept LoRAs.
* Using inpainting checkpoints (like DreamShaper Inpainting) to swap faces on pre-existing images.

Original YouTube Tutorial: https://youtu.be/aJwl7z-1iB8

Connected Videos in this Series:
* How to Load and Stack Multiple LoRAs in ComfyUI: https://youtu.be/NYVqcA9JvJQ
* Save Disk Space! Share Models and ControlNets with Forge: https://youtu.be/JpqMOI8QlSo
* Swap Face from images with Roop: https://youtu.be/093xKY7quEc

Source/Model Resources:
* IP-Adapter FaceID Plus V2 Models (by h94): https://huggingface.co/h94/IP-Adapter-FaceID
* Stable Diffusion WebUI Forge GitHub: https://github.com/lllyasviel/stable-diffusion-webui-forge
* Automatic1111 WebUI GitHub: https://github.com/AUTOMATIC1111/stable-diffusion-webui

Video Details:
* Original Publish Date: June 21, 2024
* Focus: Character Consistency / ControlNet IP-Adapter / Face Modeling
* Requirements: Local Windows installation, updated ControlNet extension, InsightFace dependency.

Follow lordcaocao2025 on Dailymotion for more deep-dives into advanced ControlNet workflows, high-fidelity character creation, and local hardware optimization!

---
Connect with me:
📺 YouTube: https://www.youtube.com/@CaoCao2025
📱 TikTok: https://www.tiktok.com/@caocao20250
💎 Patreon: https://www.patreon.com/cw/Caocao2025

#StableDiffusion #ControlNet #IPAdapter #FaceIDPlusV2 #CharacterConsistency #Automatic1111 #WebUIForge #lordcaocao2025
Transcript
00:00Hello, guys. Welcome back with me, Chow Chow 2025.
00:04Today, we're going to learn how to use any face for your image using ControlNet.
00:09The good news is, unlike Roop, it is work for any model, including realistic cartoon, 3D, or animai model.
00:19It is faster than Roop, and you could also use it with Forge.
00:23And it is also uncensored.
00:26Like Roop, you don't really need any LoRa training for this.
00:31All you need is ControlNet.
00:33So that is why not only Automatic 1111 Stable Diffusion that could use this,
00:39ForgeWeb, OUI, and Comfy also could use this.
00:43Unlike Roop, who changed the face after the image is finished, in Face ID plus V2,
00:49the faces work when you generate image, so that is why it could work for any model.
00:54And you could put animai model on realistic picture, and vice versa.
01:00Also, it support SD 1.5 and SDXL.
01:14Now let's start by updating your Automatic 1111 or Forge or Comfy.
01:19In your Forge or Automatic 11, 11 or ComfyU, you need just double-click update.bat.
01:31After that run, like usual.
01:34If you use Automatic 1111, then there is extra step to update the ControlNet extension.
01:40Just go to the extension, click check for update, and search if your ControlNet need to update.
01:46If that's the case, click update, and then apply and restart you.
01:51The next step is to download the model.
01:55Go to huggingface.co.h94.ip-adapter-faceid.
02:03Then, from there, go to Files and Version.
02:07You need to download.
02:09IP-adapter-faceid plus v2-sd1.5.bin if you use SD.5 model or
02:17IP-adapter-faceid plus v2-sdxl if you use SDXL model.
02:25Put those two file in your ControlNet model location.
02:29The default location is at webui-models-controlnet.
02:36Next, you also need to download the LoRa model that is IP-adapter-faceid plus v2-sd1.5 LoRa.safetensors.
02:45If you use SD.1.5 model or IP-adapter-faceid plus v2-sdxl LoRa.safetensors.
02:57If you use SDXL model.
03:01Put those two file in your LoRa model location.
03:05The default location is at webui-models-loRa.
03:10After that, you refresh your ControlNet and LoRa model list.
03:14After that, select IP-adapter-in-control type, select InSiteFace plus ClipH in preprocessor,
03:24and select IP-adapter-faceid plus v2-sd1.5.bin as model.
03:32Lastly, use LoRa and select IP-adapter-faceid plus v2-sd1.5 LoRa.
03:41You could use it from drop-down list or simply write it on the prompt like regular LoRa.
03:47Next, you need to search the picture that you want to use.
03:50Try not to use picture that not have the full face or wearing certain accessories like hijab, hat, or black
03:57glasses.
03:58There is a chance that the model not recognize it as the face.
04:01And also, you need to put a picture with a single person obviously.
04:07For example, I use this guy picture.
04:10Just simply add prompt like regular.
04:13Now generate this, and it will work like a charm.
04:33You could also edit the control net, like when it start or stop using the control net,
04:39as reference to gave the generation an adjustment so it not look really strict.
05:07And don't worry, you could also still use other LoRa for this.
05:11For example, I use Indonesian high school uniform LoRa for this.
05:39In case for changing existing image, you need to use in painting for this.
05:45In case for changing existing image, you need to use in painting for this.
05:47Finish Roopit changing face.
05:48You need to play with the prompt, the control net setting like model wait, and start and stop.
05:55The In Painting model also really help here, so make sure to use one like dream shaper In Painting.
Comments
lordcaocao2025
Creator
Have you had a chance to test this out against standard LoRA training yet? Let me know below!

Recommended