1 Click Automatic Installer & A Very Advanced GUI For InstantId For Windows, RunPod, Linux & Kaggle Notebook

This post is now only history and obsolete

new post : https://www.patreon.com/posts/118469722

This post is now only history and obsolete

new post : https://www.patreon.com/posts/118469722

This post is now only history and obsolete

new post : https://www.patreon.com/posts/118469722

Join discord and tell me your discord username to get a special rank : SECourses Discord

How to use NGROK to connect Gradio apps on free Kaggle notebooks : https://youtu.be/iBT6rhH0Fjs

16 August 2024 Update:

• InstantID updated to V22

• Test all LoRA combinations added that you can test every model + every LoRA combination

• LoRA scale was not working properly when you change them - this issue fixed

• LoRA loading and unloading was not working properly - this issue fixed

• Reloading pipe in each case fixed - now only LoRA weights loaded and unloaded

• Way more accurately and faster working than before

• New Pro Headshot preset added

• Diffusers version upgraded to 0.30.0 so you should windows update and call install bat file again or manually upgrade diffusers ver

3 July 2024 Update V14:

• Zip file updated to : instant_id_v19.zip

• New feature Change applied LoRA Scale added under LoRA selection section

• You can now set the applied power / strength / scale of LoRAs

• Just run Windows_Update_Version.bat if you had Version 13

• Libraries updated to latest versions

• Per step speed added to the command line interface window (CMD - Terminal)

• Kaggle notebook updated to V4 - now uses kaggle/temp directory so we can download more models

• A cell added to the Kaggle notebook to download SDXL Base 1.0, RealVisXL 4.0, Juggernaut_XL_v9 and Juggernaut_XL_v10 into models folder automatically

30 April 2024 Update V13:

• InstantID updated to V13

• FP4 and FP8 support added. Thus even with 8 GB GPUs now you can use amazingly

• Fully supporting older weak GPUs as well

• Kaggle notebook updated and now works much better than before with FP8

• Kaggle notebook is inside the zip file

• Massed Compute installers and instructions added to the zip file

• RunPod installers and instructions added to the zip file

• Newest lower precision and cpu offloading required VRAMs as below

• Several metadata saving and loading bugs fixed. E.g. it was saving style prompts as well or was giving error when you didn't use any

18 March 2024 Update V12:

• InstantID updated to V12

• Now Metadata of the images will be saved into saved images

• You can load and see Metadata from newly added Image Metadata tab

• Load & Set Metadata Settings buttons will load and set the same settings of particular generated image including input images

• If images are not existing anymore they won't be set

• update_windows_version .bat file will update to this version if you have V11

• If still doesn't work try to run update_windows_requirements .bat

• If still not work install into a new folder like C:/instant_id_v13

25 February 2024 Update V11: 

• LoRA model support added

• You can even use multiple LoRAs at the same time 

• Select the LoRA models from LoRA model dropdown 

• The Gradio will scan loras folder inside the InstantID folder

• You can also give custom LoRA models folder path with this argument --loras_path 

• Example full argument 

• python gradio_demo\web-ui-multicontrolnet.py %extra_args% --models_path "G:\automatic1111\stable-diffusion-webui\models\Stable-diffusion" --loras_path "G:\automatic1111\stable-diffusion-webui\models\Lora"

• Download latest InstantID and extract into folder where you want to install

• LCM sampler bug fixed but it is really bad quality

• Installer library version bug fixed

• App screenshot updated : instant id app screenshot.png

21 February 2024 Update:

• Face cropping feature massively improved to support

• Different aspect ratio having input face

• Different aspect ratio having outputs 

• Please do a fresh install as we have a new library dependency 

• Also the detected face and cropped input images will be saved inside temp_faces folder inside InstantID foder

• We also did a fix to lowvram option which was causing 1 click working 1 click error 1 click working

• Please report any other errors you might encounter 

20 February 2024 Massive Update:

• Fixed model reloading with V5 update

• New style and depth model added as well

• In rare occasions CPU Offloading could cause error especially on Kaggle

• This error is fixed

• Added refresh models feature - Only SDXL models work

• Fixed RunPod instructions error

• If you want to use existing models folder: edit run_web_UI. bat and add your folder path as below  

• --models_path "G:\automatic1111\stable-diffusion-webui\models\Stable-diffusion"

• Don't forget "" and give your own model path

18 February 2024 Massive Update:

• Now supports Pose, Canny & Depth ControlNets - all auto installed & downloaded

• Single controlnet web app removed since all optimizations applied to multi controlnet and should work very well even with 8GB GPUs

• When starting the app it will ask you Low VRAM option, Gradio Live share option and FP16 option

• Use Low VRAM option if you have lesser than 16 GB GPUs

• Low VRAM option will enable CPU offloading and it is super optimized. All thanks to amazing work of Pelolisu from our Discord channel

• Use FP16 option if you have older GPUs such as RTX 1080 TI which doesn't support BF16

• Use Gradio Live share if you want to share your app or if you are running it in RunPod or such cloud, etc

• Kaggle notebook also updated to latest version

• On Kaggle notebook, if you want to use ControlNets you need to add --lowvram to starting parameter

• However this will slow down the generation a lot

• Without additional ControlNets, Kaggle works super fast

• RunPod instructions txt file added. Just change the paths if you are a Linux user

10 February 2024 Update:

• Automatic image cropping feature added so now you can upload any aspect ratio image and get any aspect ratio output image - it is auto active

• CPU offloading feature added to the single controlnet gradio app so it works super fast with 12GB / 10GB GPUs even 8GB gpus should work fine now just reduce target resolution

• Multi-ControlNet app is also now supporting CPU offloading and with 16GB GPU you should be able to use it very fast with multiple ControlNet at the same time

• To update either do a git pull or do a fresh install

• How to do git pull : https://youtu.be/-NjNy7afOQ0

Full tutorial video : in production

Fully automatic installers for Windows and Linux for InstantID : Zero-shot Identity-Preserving Generation in Seconds > https://instantid.github.io/

Watch it to learn how to install Python (3.10), Visual C++ tools properly > https://youtu.be/-NjNy7afOQ0 since you need C++ tools installed for InsightFace library

Features

• 1 click install on Windows (local) and RunPod and Kaggle (free account) and Linux (local or cloud) systems

• It will install everything and download everything into a new VENV

• Advanved Gradio Web APP compared to the official demo with following features

• xFormers - faster

• Custom image resolutions - enter any final resolution of image you want (some specific resolutions may fail but common ones working like 768x1024, 1024x1280, 768x768, etc.)

• Scans models folder and let you pick any model from there

• Give full path or Hugging Face repo of any model with an input box

• Generate any number of images like set 200 images and it will generate 200 image

• Every image is saved by default into the outputs folder

• I am talking with developers to add extra lower VRAM (probably below 6 GB) feature

• Supports pose, canny & depth ControlNet combination

Comparisons & Experiments

• CFG comparisons 4 vs 4.5 vs 5, 6, 7, 8, 9, 10 (I think 4 is best with our technique) :

• https://imgsli.com/Mjc2ODcw/0/1

• Schedulers comparison (I think DPMSolverMultistepScheduler-Karras is best) :

• https://imgsli.com/Mjc2ODcy/0/4

• IdentityNet strength comparisons (I feel like 0.6 is best with our technique)

• https://imgsli.com/Mjc2ODc5/5/3

• Image adapter strength comparisons (I feel like 1.0 is best with our technique )

• https://imgsli.com/Mjc2ODg0/7/5

Tips

• Select target resolution with same aspect ratio of your input image

• Crop your input image with same aspect ratio of your desired output resolution

• E.g. input image is 1746x2619 so target resolution can be 768x1152 - same aspect ratio 

• I use resize feature of Paint .Net (free open source) to find target resolution or cropping 

• You can download any SDXL model from CivitAI or Hugging Face and use. Download into models folder inside InstantID folder and restart the APP

Requirements

• Tested on Python 3.10.11 so use Python 3.10.x

• C++ tools has to be installed for InsightFace library

• Full tutorial to install these 2 : https://youtu.be/-NjNy7afOQ0

Windows Install Instructions

• Download instant_id . zip from attachments

• Extract files into any folder you want to install

• First double click windows_install .bat and install the InstantID

Kaggle Instructions

• Upload kaggle notebook inside zip file (import notebook)

• Select Python and GPT T4 x2 and Internet On

• Instructions are well written on the Notebook