v4.0.0

[C0untFloyd]

Bugfix: Unnecessary init every frame in live-cam
Bugfix: Installer downloading insightface package each run
Added xseg masking to live-cam
Added realesrganx2 to frame processors
Upgraded some requirements
Added subtypes and different model support to frame processors
Allow frame processors to change resolutions of videos
Different OpenCV Cap for MacOS Virtual Cam
Added complete frame processing to extras tab
Colorize, upscale and misc filters added

This release adds complete frame processing functions like upscaling and colorizing to the extras tab.

Please note: to get rid of the insightface download every time using the automatic installer replace your old windows_run.bat with the new one from the installer folder.

---

This discussion was created from the release v4.0.0.

[lysxelapsed]

Nice update! File upload progress is fixed too 👍 Will test the new extras later.
As always, thx for the amazing work!

[Pr0f3ssi0n4ln00b]

Gonna wait a bit for potential bugs to get fixed, as I have no issues with my current install. What do you mean new models in assets? Can I (and should I?) download just the models in question?

[C0untFloyd]

New models for colorizing, upscaling etc. (the frame.zip in assets). Either download the archive or have roop download them at start. It doesn't matter.

[RoopUser]

I can't quantify this (because "feelings") but things just doing standard jobs seem to produce less accurate output as of the last update (no post processing, retry rotated). Seems I am getting more flickering and distorted attempts on frames (even with some slightly-off coloring).

I also noticed the "Number of swapping steps" is out in the main section. I have tried with 2, and again with 4. Does this affect things much?

Has anyone else noticed anything similar?

[RoopUser]

In fact, doing anything >1 number of steps makes the swapped face seem more saturated/tinted looking. Doing so is just worse overall for results and performance it appears.

[lysxelapsed]

I think that's a given. While more than one swapping step can increase likeness, it will also increase face coloring, visibility of the "box" etc. So it really depends on the source (mostly) and target face. Maybe increasing erosion iterations could help? Never tried it myself.
In the end it's a matter of opinion / taste / choice. I do have some input faces, where increasing swapping steps gives a worse result. But safe to say that 2 swapping steps is my new standard. More than 2 is a bit overkill for me and looks unnatural in most cases.

[HansM65]

Hi just tested the 4.0. It was dowloading all the updates automatically. Using it on a batch of images to see if the quality is different, it stays with 0% progress, but its doing the work in the backgorund, But then its stops somewhere along the line (like 5 images or so) and never finishes so i have to shut it down and restart. Thanks for the great work so far, just to let you know.

[Jiwes]

Does anybody know how to change output format for saving preview images ? it was jpg .. now it saves in webp format.

Where is setting for that ?
How to change that ?

[C0untFloyd]

#693

[Jiwes]

i use chrome and windows

[LucianoDaluz]

Thanks! It is working great and faster!

[etruscanmuffin909]

This. It's incredible

[lysxelapsed]

Did a quick test with esrgan x2 upscaling. I think it's only suitable for short vids and fast cards. With the RTX3060 @ 70% GPU load, it took somewhere around 1-2 seconds per frame. For comparison: 4x upscaling with Topaz Video AI (Proteus4 model) generates 8-12 frames per second.

[CassySpencer]

Forgive the totally newbie question here but how do I install this on my Mac?

[Nkoooz]

how to do ????

[rockzedge]

I am restoring Nippon Showa old videos.
This is professional singer Asaka Yui (18y old in those days) ”Believe Again” , her 8th EP vinyl, 1988.

https://youtu.be/9HGme3cay8M?si=CDdvgHgEQN1PmziK

First of all, upscaling with Topaz video AI. (480i to 1080p Iris3 )
and then roop, face picture is using stable dufusion forge. (with original Yui's-LORA)
Usually her mouth is overing the microphone in previous versions,
New masking processor makes cool job. (...although there is a slight collapse of the metal parts.)

And I can't help but say it, Please do not pollute this thread with silly questions, I just know thousand of try and error makes you happy.

I love this Fork of Roop, with deepest gratitude.
Thank you very much C0untFloyd-san.

From Nippon otaku user

PS. Guys, I wish to know what you are making with roop.

[LucianoDaluz]

So you are using Roop inside Forge? Can you share the install link? Thanks

[rockzedge]

Sorry for the lack of explanation.
It's not a SD-Forge extension, but a standalone one.

[LucianoDaluz]

Sorry for the lack of explanation. It's not a SD-Forge extension, but a standalone one.

How did you use a LoRa on it?

[rockzedge]

The model is trained using 30 photos of YUI.
https://github.com/hako-mikan/sd-webui-traintrain

[LucianoDaluz]

I know. What I thought is that you used a LoRA inside Roop.

[ali0une]

This is a major version. Many thanks for your hard work, much appreciated!

[Nkoooz]

I still can't do it, do you have an explanatory video? Le lun. 27 mai 2024 à 09:12, rockzedge ***@***.***> a écrit :

…

Excuse me but did you read the link below? https://github.com/C0untFloyd/roop-unleashed/wiki/Installation-:-MacOS — Reply to this email directly, view it on GitHub <#682 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BIWZ3E4KQ7IULCGRRJXE6STZELMFFAVCNFSM6AAAAABHTB7RS2VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TKNRWHE4DM> . You are receiving this because you commented.Message ID: ***@***.*** com>

[rockzedge]

Hello Nkoooz-san
First of all, I don't think this is the place to ask questions, please delete quetions and make a new thread.

I'm not an engineer, but I can Google it.
Please refer bellow.
Nike-Inc/gimme-aws-creds#351

[jarrodmedrano]

Do not click that link it contains a virus ^

[Shahfahad7866]

Which option is better, Windows 1 Click Installer or Manual? I Got Cuda 12 installed and using RTX 4070 with 12 GB of VRAM.
I am not sure, how to install this file, which I downloaded from: https://developer.nvidia.com/rdp/cudnn-archive

[awpeoith1254]

Hi - couple things:

• You need to get Cuda V11.8 unless something has changed, not V12.
• For CUDNN just need to copy those files into where you installed GPU Computing Toolkit
• So for me it's: Program Files --> Nvidia GPU Computing Toolkit --> CUDA --> V11.8. In that V11.8 folder you will see the same folders as CUDNN (e.g., bin, lib).

Hope that helps!

[Shahfahad7866]

Thanks a lot sir.

[jarrodmedrano]

Why did you remove text masking? Add it back please.

[etruscanmuffin909]

#788