InstantID Cog Model

Overview

This repository contains the implementation of InstantID as a Cog model.

Using Cog allows any users with a GPU to run the model locally easily, without the hassle of downloading weights, installing libraries, or managing CUDA versions. Everything just works.

Development

To push your own fork of InstantID to Replicate, follow the Model Pushing Guide.

Basic Usage

To make predictions using the model, execute the following command from the root of this project:

Note:
default SDXL model: AlbedoBase XL V2
default scheduler: 4-step sdxl-lighting for fast inference

cog predict \
-i face_image_path=@examples/halle-berry.jpeg \
-i prompt="woman as elven princess, with blue sheen dress" \
-i negative_prompt="nsfw" \
-i adapter_strength_ratio=0.8 \
-i identitynet_strength_ratio=0.8 \
-i safety_checker=True

To change the denoising steps, use argument:

-i lightning_steps="2step" (or "8step")

To use a custom scheduler, pose controlnet and a different base SDXL model:

Example:
cog predict \
-i face_image_path=@examples/halle-berry.jpeg \
-i pose_image_path=@examples/poses/ballet-pose.jpg \
-i prompt="photo of a ballerina on stage" \
-i model="Juggernaut XL V8" \
-i adapter_strength_ratio=0.8 \
-i identitynet_strength_ratio=0.8 \
-i pose=True \
-i pose_strength=0.4 \
-i enable_fast_mode=False \
-i scheduler="DPMSolverMultistepScheduler-Karras" \
-i num_steps=30 \
-i guidance_scale=4 \
-i safety_checker=True

Input Parameters

The following table provides details about each input parameter for the predict function: