Consistent AI Characters with PuLID and FLUX GGUF in ComfyUI: No LoRA Needed!

 


We're exploring the world of PuLID, a revolutionary tool that will transform how you create consistent characters in your AI art.  If you've struggled with maintaining the same look and feel across multiple AI-generated images, then PuLID is your answer.  Even better, we'll be using the incredibly efficient Flux gguf model, meaning you can achieve fantastic results even with limited VRAM.  Let's get started!


What is PuLID?

PuLID stands for "Personalised Latent Image Diffusion." It's an innovative approach to AI image generation that focuses on preserving character consistency without relying on traditional fine-tuning methods like LoRA (Low-Rank Adaptation) training.  In simpler terms, PuLID allows you to make edits and adjustments to your AI-generated characters while retaining their core features, style, and composition.

If you prefer to watch the tutorial video, watch here:



Why Use PuLID with Flux?

Flux gguf is a diffusion model known for its high-quality image generation and impressive efficiency on both CPUs and GPUs.  It requires as little as 4GB of VRAM, making it accessible to a wider range of users.  Combining PuLID with Flux is a powerful combination for achieving consistent, high-quality character generations without the extensive resources often needed for training.

Setting Up PuLID in ComfyUI

  1. Install ComfyUI and Manager Node:
    If you are new to ComfyUI, follow our comprehensive [ComfyUI Installation Guide](link to guide). Ensure you have the ComfyUI Manager node installed.  This is essential for downloading and managing all the models required for our workflow. For those opting for premium one-click installations, you are all set.

  2. Download the PuLID Model: The PuLID Flux model isn't currently available through the ComfyUI Manager, so we'll download it directly from Hugging Face. You can find the model file here: [Hugging Face PuLID Flux Model](link to model).

  3. Place the PuLID Model: Once downloaded, create a new folder named "PuLID" within your ComfyUI "models" directory. This is where the downloaded PuLID Flux model will reside. 

  4. Install Custom Nodes: Now, launch ComfyUI and load the basic workflow provided in the description below. Don't be alarmed if you see red nodes.  This simply means that the required custom nodes haven't been installed yet. Open the ComfyUI Manager, click "Install Missing Nodes," and install each one individually. Restart ComfyUI to ensure the changes take effect.

  5. Gather Remaining Models: Before we launch ComfyUI, we need to gather the other essential models.  For this workflow, you will need.

  • Flux gguf Model: Search for "Flux gguf" in the ComfyUI Manager's Model tab and install your preferred version (either Dev or Schnell Quantized).
  • CLIP Models: Search for "clip" and install the T5xxL FP8 and the Comfy Anonymous CLIP L. 
  • VAE Model: Search for "Flux vae" and install the Flux.1 model.
  • InsightFace Models: Search for "insightface" and install all five of the Antelope V2 models.


Using PuLID

  1. Load Image and Prompt:  Upload your base image – the one you want to transform – and type your prompt in the "CLIP Text Encode (Prompt)" node.  Describe the changes you'd like to see, like "a woman eating strawberries."

  2. Queue and Generate: Click "Queue Prompt" and let PuLID do its magic!  Be patient, as generations involving face analysis can take a little longer than usual.

  3. Explore and Refine: Play around with different prompts, images, and settings to unlock PuLID's full potential. You will be amazed at the level of control and consistency you can achieve.


Conclusion

PuLID is an amazing tool for consistent character creation in AI art.  Combined with the efficiency of Flux gguf, it empowers artists of all levels to generate high-quality images without needing advanced technical expertise or expensive hardware. I encourage you to explore this exciting tool and unlock a new level of creativity in your AI art journey.


ComfyUI Github - https://github.com/comfyanonymous/ComfyUI

ComfyUI Manager Github - https://github.com/ltdrdata/ComfyUI-Manager

PuLID Model - https://huggingface.co/guozinan/PuLID/tree/main



ComfyUI Basic Workflow:

https://drive.google.com/file/d/1-lFZ5KSnOmxm5KortODh8SieyItNJTrO/view?usp=sharing


Commands for install:

.\python_embeded\python.exe -m pip install -r .\ComfyUI\custom_nodes\ComfyUI-Manager\requirements.txt

pip install -r .\ComfyUI\custom_nodes\ComfyUI-Manager\requirements.txt



Post a Comment

2 Comments

  1. This comment has been removed by the author.

    ReplyDelete
  2. PuLID is a game-changer for creating consistent AI characters without LoRA. With MimicPC's ready-to-use workflows featuring PuLID and Flux , you can easily generate stunning images with advanced features like one-click face-swapping. Perfect for artists seeking efficiency and quality!

    ReplyDelete