EchoMimic Magic: Audio and Landmarks Bring Portraits to Life! The BEST talking head generation app.

Readme / Instructions
drive.google.c...
#AI #StableDiffusion #TechInnovation #ArtificialIntelligence #DeepLearning #AIExploration #TechEnthusiast #CreativityInAI #StableAIHub #AICommunity #InnovationHub #TechBreakthroughs #AIResearch #futuristictechnology #EchoMimic #audiotovideo #lipsync

[@StableAIHub]

BAT file for launching @echo off REM Change to the directory of the batch file cd /d "%~dp0" REM Activate the EchoMimic environment call conda activate echomimic REM Launch WebUI python webgui.py --server_port=3000

[@IdgrafixCh]

Hi there, I tried ton install it for ComfyUI, but could not do so successfully (via Manager + copy paste all missing files from the repo). Could you please make a tutorial for ComfyUI please? 😊

[@StableAIHub]

@@IdgrafixCh I am sorry, Comfy is not my cup of tea. It is a standalone install why use Comfyi. Comfy will occupy VRAM for it's own use and then this tool. Standalone means more VRAM available. That's my understanding.

[@Im_that_guy_man]

I just wanna say thank you for your tutorials. great job

[@StableAIHub]

Thank you for your feedback

[@arron122]

👀Gonna test this one out

[@ChikadorangFrog]

This AI is just the best out there in terms of quality compared to all others (Hedra, Liveportrait, Sadtalker, Hallo, V-express). It might be the right time to start saving for RTX 5090

[@StableAIHub]

Ha ha ha. True that, let me also start saving. Please could you check the teeth part. Are you happy?

[@StableAIHub]

I think eye blinking needs some improvement. Sometime only 1 eye blink.

[@ChikadorangFrog]

@@StableAIHub Minor imperfections are ok. It's easy to edit in capcut by applying some effects. What is important for me is the skin texture similar to sadtalker.

[@StableAIHub]

@@ChikadorangFrog Right. The quality is good. I wasn't expecting this good for AI.

[@madmad555]

the auto lip sync is cap tho, better in sadtalker

[@behrampatel4872]

hi this has come up in another youtubers video but is it necessary to use conda to create virtual environment's ? From all your videos i learned that we can create a venv on our own. So will this tutorial work if we don't use conda ? Thanks, b

[@StableAIHub]

The answer is long. Primarily we are using either PIP or CONDA to create virtual environment (VE). Sometimes the dependencies are very specific like some packages, python version... etc which can be easily done using conda. I don't know if this will work without conda. You need to try and let us know plz.

[@behrampatel4872]

@@StableAIHub Got it. thanks for the info. Cheers, b

[@Avalon19511]

Thank you for the video, unfortunately I think the big roadblock with a lot of these talking head software is optimization, it took 17 minutes for a 5 second video, imagine if you had a 3 minute video, it would take 6hrs and 12 minutes which is just not a good use of your time, hopefully in the neat future they get better

[@StableAIHub]

The processing time can be significantly reduced if you use a 16GB or 24GB VRAM card. Using cloud services can further decrease the rendering time. A few months ago, the major issue was the quality, as the output would get distorted when using realistic images generated with SD. EchoMimic has surprised me with its improvements. I'm happy to see that the quality is getting better, and in due time, the speed will also improve. Unfortunately, I have the most basic laptop that only meets the minimum requirements, which explains the slow speed.

[@Avalon19511]

@@StableAIHub I think you'll agree, prices being what they are, most people either have a 12 or 8 gb and I think that is where the optimization focus should be:)

[@StableAIHub]

I agree what you said. I hope in due time the processing would be much faster on low VRAM cards.

[@ChikadorangFrog]

@@StableAIHub i think devs plan to release a faster version of this in 1 to 2 months

[@TomiTom1234]

Good tool, better than HALLO which takes longer time to process. BTW, I created a bat file to start the program easier and faster.

[@ChikadorangFrog]

can you share the bat file?

[@TomiTom1234]

@@ChikadorangFrog The video publisher added the code for bat file and pinned it, you can copy and paste it in a text file, then change the extension to "bat".

[@ChikadorangFrog]

@@TomiTom1234 thx

[@TomiTom1234]

@@ChikadorangFrog You are welcome. Don't forget to change the paths that need to be changed to match your folders.

[@VintageForYou]

I have installed EchoMimic when I load an example image and audio I get an Error can you please help.🤔 Error code,,, cv2.error: OpenCV(4.10.0) :-1: error: (-5:Bad argument) in function 'resize' > - src is not a numerical tuple > - Expected Ptr for argument 'src'

[@StableAIHub]

Check if the solution posted here works? github.com/BadToBest/EchoMimic/issues/102

[@VintageForYou]

@@StableAIHub Got it working now from your link but it takes time to render for 5 Seconds of audio on a 12GB Graphics card and 32 GB of RAM over 20 minutes this app is similar to Hallo Time consuming.😥

[@StableAIHub]

@@VintageForYou Try the accelerated version which is very fast.

[@Nonewedone]

I got the same issue.

[@Nonewedone]

File "webgui.py", line 169, in process_video face_img = cv2.resize(face_img, (width, height)) cv2.error: OpenCV(4.10.0) :-1: error: (-5:Bad argument) in function 'resize' > Overload resolution failed: > - src is not a numerical tuple > - Expected Ptr for argument 'src'

[@ChikadorangFrog]

Is the new update working? Im having lots of errors

[@StableAIHub]

A2V with acceleration is working fine. Please could you share error screen using Drive.

[@ChikadorangFrog]

@@StableAIHub Thanks its working fine now. The Gradio is the one that is not working

[@StableAIHub]

@@ChikadorangFrog If no one is gonna fix I will see if I can. I am not a programmer so gonna take help from AI. By any chance do you have the old version / earlier release of EchoMimic when it was working

[@StableAIHub]

@@ChikadorangFrog Please check the github, I posted the solution. If you can confirm on github, it can be merged in repo

[@ChikadorangFrog]

@@StableAIHub i made a mistake by cloning the latest version and copy paste it to the original/old. I no longer have the old working version

[@ChikadorangFrog]

The quallity of the accelerated version is not good. I will just use the slower version for now

[@StableAIHub]

I noticed the same. Used the slower version for next video. Did you came across any tool for singing talking head.

[@ChikadorangFrog]

@@StableAIHub next release of echomimic would have Pretrained models with better sing performance to be released

[@StableAIHub]

We need to keep a watch on ingrid789.github.io/MyTalk/ Looks amazing

[@ChikadorangFrog]

@@StableAIHub might be good to combine with Kling AI

[@user-fo9ce3hr5h]

29 second audio for 2 hours? my gpu 4070 super. wow, what is your gpu sir? this is so long. maybe we should go for something faster. btw thank you.

[@StableAIHub]

You can try accelerated version which is much faster but quality is less. Also try SadTalker which is very fast kzbin.info/www/bejne/l3PTkmiph9OIfZY

[@user-fo9ce3hr5h]

@@StableAIHub Thank you so much brother i will install one of these. i hope i can make an avatar with these. thank you.

[@rahulkathuria8250]

output video isn't HD, blurry

[@StableAIHub]

It is trained on 512 x 512 dataset. Use upscaler to improve quality.

[@StableAIHub]

I always use 4xUltraSharp in Automatic1111. For that you need to extract all frames, upscale and then combine as video. You can refer the following on how to extract frames kzbin.info/www/bejne/aH6Zg3ZnoK-Yn9E

[@rahulkathuria8250]

@@StableAIHub beard is getting blurry and distorted

[@StableAIHub]

@@rahulkathuria8250 Do you have generated video. Please post on github

[@rahulkathuria8250]

beard is getting blurry and distorted

[@StableAIHub]

Please post the output on github

[@rahulkathuria8250]

@@StableAIHub you mean the video, okay but they haven't released the dataset which means they haven't trained bearded guys.