Install and Use Infinite-You in ComfyUI: Complete Setup Guide

In this how to searies, you'll learn how to installing and using Infinite-You in ComfyUI—a powerful tool to create multiple images of yourself, others, pets, or AI characters from a single reference image.

We’ll cover everything from setup to advanced workflows, including using Promptus Studio Comfy (PSC), a leading platform built on the open-source ComfyUI framework.

🤖 What Is Promptus Studio Comfy (PSC)?

Before diving into installation, understand PSC’s role:

• Browser-Based & Cloud-Powered: Access ComfyUI workflows via CosyFlows (no-code interface), real-time collaboration, and distributed GPU compute for faster, high-res outputs.
• Advanced Model Access: Built-in integration with models like Gemini Flash, HiDream, Hunyuan3D, Stable Diffusion, GPT-4o, and Gemini.
• Workflow Publishing & Discord Integration: Share and collaborate easily, making it popular among creative teams and solo creators.
• Multi-Modal Support: Generate text, image, and video within one platform.
• Ease of Use: Ideal for users who want ComfyUI’s power without deep technical complexity.

⚙️ Setting Up Your Environment

You can follow this tutorial on RunPod or locally (Windows, Linux, or Mac). The folder structure and steps remain consistent.

🖥️ RunPod Installation Process

1. Access Template: Use the provided RunPod template link to auto-set up ComfyUI.
2. Select GPU: For this guide, we use an A100 PCIe with 80GB VRAM—adjust based on availability.
3. Configure Environment Variables:
   • In Public Environment Variables, set required parameters (e.g., LoRA Depth, T5 Clip Model, additional Infinite-You models) to “1” for automatic downloads.
4. Launch Services: Once provisioned, you’ll see:
   • Jupyter Lab: File manager for uploads or manual edits.
   • ComfyUI Interface: Visual workflow editor to build Infinite-You pipelines. ⚡

🗂️ Initial Setup and File Structure

• Folder Layout: The template or local repo includes folders like models/checkpoints, custom_nodes, etc.
• Accessing Files: Use Jupyter Lab (on RunPod) or your local file explorer to verify folders.
• Workflow Editor: Open ComfyUI at its localhost URL (e.g., localhost:8188) to build or import workflows.

🔧 Installing Infinite-You Nodes

1. Open ComfyUI Manager

• From the main interface, launch the Custom Nodes Manager.

2. Search & Install

• Search: Enter “Infinite-You” in Custom Nodes.
• Select Version: Choose “Infinite-You Human” (or relevant variant).
• Install: Click install; wait ~2–3 minutes.
• Try Fix: If installation fails, use “Try Fix” to reinstall problematic nodes.

3. Restart Interface

• After installation, restart ComfyUI when prompted and refresh the browser. 🎉

🔍 Understanding the Workflow Components

The Infinite-You pipeline involves several key elements:

1. Model Configuration

• Diffusion Model:
  • Uses Flux-1-dev with FP8 E4M3 weights by default.
  • Optionally swap to quantized versions to reduce GPU requirements.
• Load Node: Ensure the “Load Diffusion Model” node points to the correct model file.

2. Dual Clip Loader

• Clip Models: Requires T5-XXL-FP8 and Clip-L models.
• Formats: Supports FP16 and FP8 for memory optimization.
• Type: Set to “Flux” or matching the chosen diffusion model.

3. Text Encoding Setup

• Separate Inputs: Two text boxes (one for FP16, one for Clip-L) allow split prompting for varied results.
• Negative Prompts: Typically left empty for Flux-based workflows.

4. ID Embedding Configuration

• Automatic Download: The workflow auto-downloads required embedding models.
• Stages:
  • Stage 1: Less creative—follows input face closely.
  • Stage 2: More creative—allows divergence from the reference.
• Provider Settings:
  • Face Analysis Provider: “Cuda”
  • Face Analysis Size: “auto”
  • Image Projection Tokens: Set to 8
• Model Variants:
  • SIM Stage 1 & Stage 2 in FP16, BF16, or other precisions—choose based on GPU memory.

⚙️ Advanced Features and Controls

🧩 Face Control Net (Experimental)

• Current Status: Included but not fully functional for pose control.
• Intended Workflow:
  1. Load a control net pose or reference image.
  2. Extract face pose via face detector.
  3. Apply pose constraints to generated images.
• Workaround: Leave pose image input empty to avoid conflicts until feature matures.

🔄 Optimization Parameters

• Strength: Leave at 1.0 for standard likeness.
• Start Percent: Try ~0.10 for subtle guidance early.
• End Percent: Try ~0.90 to retain likeness while allowing changes.
• Tips: Adjust these to balance fidelity vs. creativity and experiment per reference.

▶️ Running Your First Generation

1. Load Reference Image: Connect your single reference (face/person/pet) to the designated input node.
2. Enter Prompts: Fill both text encoding boxes with your desired style/context prompts.
3. Verify Models: Ensure diffusion, clip, and embedding models are loaded without errors.
4. Queue Prompt: Click “Queue Prompt” (or equivalent) to start.
5. Monitor Progress: Watch logs or previews in ComfyUI; adjust parameters if needed. 🎨

📈 Results and Quality

• Likeness Transfer: Infinite-You excels at maintaining facial features while adapting to new contexts/styles.
• Variations: Try Stage 1 for closer matches, Stage 2 for creative variations.
• Prompt Crafting: Use descriptive prompts (e.g., “portrait in cinematic lighting, vibrant colors”) to guide aesthetics.
• Batch Generation: Generate multiple outputs, then select favorites for further refinement or upscaling.

🌐 Getting Started with Promptus

• Sign Up: Visit https://www.promptus.ai and create an account.
• Access Options:
  • Promptus Web: Browser-based, no local setup needed.
  • Promptus App: Desktop integration for shorter latency.
• Import Workflows: Upload or recreate Infinite-You pipeline via CosyFlows (drag-and-drop).
• Distributed Compute: Leverage cloud GPUs to run heavy models without local hardware constraints.
• Collaboration: Share workflows, integrate with Discord, collaborate in real time. ✨

🛠️ Troubleshooting and Support

• Node Import Failures: Use “Try Fix” in ComfyUI Manager; ensure correct Python/environment versions.
• Model Download Issues: Check network access in RunPod or local; manually download and place model files if needed.
• GPU Memory Errors:
  • Choose lower-precision variants (FP8/FP16) or smaller models.
  • Use cloud GPUs (Promptus or RunPod) if local VRAM is insufficient.
• Interface Refresh: Always restart/reload ComfyUI after installing nodes or models.
• Community Help: Join Discord servers or forums for ComfyUI/Infinite-You—experienced users share tips and custom workflows. 💡

🎉 Conclusion

This guide provides the foundation for exploring Infinite-You’s capabilities within ComfyUI, whether on Promptus Studio Comfy or a local/RunPod instance. By following these steps:

• Install the Infinite-You custom nodes
• Configure models, embeddings, and parameters
• Run your first generation with a single reference image
• Optimize via advanced settings and batch workflows
  …you’ll be able to produce diverse, high-quality character images effortlessly.

Dive in, experiment with prompts and settings, and enjoy creating personalized images at scale! 🚀😁

Happy generating!