InstantID-IPAdapter

InstantID – IPAdapter

This guide covers the workflows shown in the InstantID ComfyUI Tutorial: identity generation, IP-Adapter style control, and cleaner face swaps using img2img and inpainting.

Guide

Learn the key steps to creating effective, consistent, and personalized AI workflows.

View Guide

Download

Access JSON files, templates, and resources to help you create workflows in your ComfyUI setup.

Download Workflow

Link

Repository, custom nodes and templates needed to properly install the workflow.

Resources & Links

Guide

The goal of this workflow is to achieve consistent character identity generation, style control using IP-Adapter, and cleaner, more natural face swaps using img2img and inpainting.

  • Consistent character identity
  • Style control using IP-Adapter
  • Cleaner face swaps

The workflow is based on Stable Diffusion XL (SDXL) together with InstantID, IP-Adapter, and ControlNet/LoRA tools for identity preservation and visual consistency.

Workflow Overview

1. Character Identity
Generate consistent characters starting from a single face reference image. InstantID analyzes the reference face and transfers facial structure, identity, and main facial traits.

This is useful for:

  • Generating the same character in different scenes
  • Creating multiple poses while preserving identity
  • Building consistent characters for storytelling or video projects
2. Face Swap
A workflow focused on replacing a face while preserving lighting, scene composition, and original image structure.

This setup uses InstantID, img2img, and denoise control.

Important note about denoise

Without inpainting
Use lower denoise values, around 0.4, to preserve the original image more faithfully.
With inpainting
Set denoise to 1.0 for maximum face reference fidelity.
3. Identity + Style
An advanced workflow combining InstantID, IP-Adapter, and CLIP Vision.

This approach gives control over character identity, artistic style, atmosphere, and overall visual composition.

Requirements

This workflow was built using SDXL.

Recommended compatible models:

  • Juggernaut XL
  • RealVisXL
  • DreamShaper XL
  • Any SDXL-compatible checkpoint

Installing the Nodes

Use ComfyUI Manager to install all required custom nodes automatically.

Required Files

Control LoRA Depth Rank256 should be placed inside:

ComfyUI/models/controlnet/

Workflow Structure

InstantID works using facial embeddings, identity recognition, and facial feature guidance.

IP-Adapter adds style control, mood, composition, and artistic consistency.

CLIP Vision improves image understanding, style transfer, and visual consistency.

Practical Tips

The workflow performs best with reference images that are:

  • Well lit
  • Front-facing or semi-front-facing
  • High quality
  • With a clearly visible face

Avoid low-quality compressed images, heavy shadows, and tiny faces inside the image.

Recommended Settings

Simple face swap
Denoise: 0.3 – 0.5
Inpainting
Denoise: 1.0
Steps
25 – 40
CFG
4 – 7

Common Problems

Deformed Face: denoise too high, poor reference image, or CFG too high.

Inconsistent Identity: small reference image, aggressive style transfer, or IP-Adapter strength too high.

Plastic Looking Face: try lower denoise, lower CFG, more steps, or a different SDXL model.

Recommended Workflow Setup

  • InstantID for identity
  • IP-Adapter for style
  • Light img2img
  • Final face inpainting

Download

You can download the workflow JSON files, resources, and related updates from the repository.

Support the Project

If you enjoy my content and want to support the project: