Відео

It's a folder you have to make, just create an empty folder named "Yolov8" in the "models" directory and put the seg models in that folder. Also you can create a symlink to your regular folder - for example: mklink /D "C:\ComfyUI\models\Yolov8" "C:\ComfyUI\models\ultralytics\segm"

It's a combination of several techniques including Live Portrait, Face Fusion, Animate Avatar, Hedra and video editing tools. Really varies based on scenario and avatar I generate

It's called WaveSpeed and it's in ComfyManager

It's been awhile since I've used it but I believe you right click the category and then there's a delete option. It may have been also in the Sidebar Settings screen.

@@GrocksterRox Thank you!

Awesome, just a few more and we'll beat Mr. Beast! :) haha every little bit helps and thanks for passing along to others!

Blend of Live Portrait, Animate Anything, Hedra and video editing tools

Did you make sure that you have CUDA selected and not "default"?

@@GrocksterRox Yes, it was selected already. Now I also added --cuda-device 0 --gpu-only and then I uninstalled onnxruntime and onnxruntime-gpu and reinstalled onnxruntime-gpu.

@toledavid89 and that resolved it or it's still an issue? If still an issue, feel free to jump on the discord and we can try to figure it out

I'm so happy this is valuable and makes you happy! Feel free to share this in the channel with others so everyone can learn.

Thanks so much, interesting on the missing items but here they are that you can install: * Mile High Styler - civitai.com/models/119246/comfyuimilehighstyler * MaraScott - github.com/MaraScott/ComfyUI_MaraScott_Nodes * You shouldn't need NF4 loader I thought I removed it from the workflow but you can do that too

@ thx for the fast response buddy will have a look in a few days and let you know :).

@Nova-ul3vv Good luck

@ Worked :)

I set up a folder symlink (essentially a pointer): mklink /D "C:\StableDiffusion\ComfyUI\models\Yolov8" "C:\StableDiffusion\ComfyUI\models\ultralytics\segm" so make sure the models are in the Yolov8 one but in order not to duplicate the models you can use the folder pointer command above (include it in your run_nvidia_gpu.bat file

It really depends on the model that you're using, I really love this project0 version 3, but the effect will vary based on your base model that you're using

The magic of AI. Haha

Yep, I had a note in the video about this, check the place where the screenshot shows how to update your security settings.

​@@GrocksterRox awesome thanks, i must have missed it first watch- i'll take a second look! just found your channel. love what you're doing- really appreciate it!

@bstuartTI I'm so glad it's been helpful for you and others feel free to share with others everyone can benefit

@@GrocksterRox def sharing- gl growing the channel. i've been learning up to this point by backwards engineering workflows- but your explanations are great. keep it up!

It looks like yes...

It's a custom blend of live avatar, Hedra, some video blending and more.

Good guess!

Yep, they are linked in the description of the video (link to CivitAi where you can download free)

i think its act one by runway, i may be wrong tho.

Not yet, but there are several free services and nodes to accomplish this including Hedra, Live Portrait, Animate Avatar and more. Good luck!

Hi there, thanks for stopping by!

100% agree, it's definitely a pain in the butt, but given there isn't an easy/effective alternative, we deal with the pain to get the reward :) Really appreciate you and your feedback!

i wish i cloud help u brother.

Worked on the first try for me. What kind of errors or troubles did you face? Maybe i can help.

It might be easy if everyone jumps into the support channel on my discord server because then we can show videos and screenshots to help resolve the issue

Sounds like you and Helga need to have a collaboration session to merge the best of both worlds :)

Definitely, I'm so glad you and others are learning from this and hopefully the entire AI community will continue to learn and grow their fun new techniques!

That's an interesting challenge! The problem we have to figure out is how to change the angles of the background even though you will easily composite your characters or elements on top. Good thinking!

I'm so glad you found them helpful!

Thanks so much, I live and die by the Mile High Styler, thanks so much for creating it! :)

Thanks so much for the great feedback and totally agree with you, there's magic science going on in the background but it definitely helps speed things up! Love the data-driven approach as well - great job!

Absolutely, anything I can do to help the community, I'll try :)

Absolutely and I'll have to check them out! Continued wishes for success!

Thanks Jay, I'm so glad you found the tips and tricks useful - keep experimenting and learning!

discord.gg/pWnectzw6Z - Looking forward to collaborating! :)

Oh totally - this (as with everything else) is another tool in the toolbelt, and there's many times using the standard method will work great. I'd definitely recommend experimenting because it really opened my eyes to the possibilities. Thanks so much for the feedback!

Love the feedback, thank you so much, and thanks so much for sharing with others!

Thank you SO much! I want to help the AI community learn this stuff as much as possible (and I'm learning all the time too of course..haha)

You got it, thanks for being a dedicated subscriber!!! :)

Thanks so much! It's a mix of methods that I vary but generally it's a mix of Hedra, Live Portrait and the new animated avatar that's mixed in video editing software

Hi there, thanks so much for the feedback! To see the Cryss monitoring for the new UI, you just need to make sure you're using the latest version of Comfy (you may need to force the update by going to the update folder outside of your Comfy folder)

Yup absolutely in fact the video is already ready for your consumption :) ua-cam.com/video/WSx74Uep590/v-deo.html Enjoy!

Thanks for the feedback, I'm not sure what you mean by writing code. Can you talk a little bit more about this?

@@GrocksterRox I just feel confused when I don't know what to do in a Node type environment. With C# or Python I can search for what I need. I have done Blender tutorials for similar Nodes and I can follow along, but I don't feel like I learn it. I just find Node environments more intimidating than Code I can just copy a working example even I don't understand it. Probably has something to do with an old dog trying to learn new tricks.

Feel free to jump on the discord and we can chat or if you need a more in depth walk through / lesson, you can sign up for 1:1 LIVE training sessions - www.grockster.com/services

Interesting. Sorry you didn't get good results. Do you want to jump on the discord and we can take a look at the settings and prompt used etc?

@@GrocksterRox Sorry :) I don't even have and ever used discord. I feel like a first grader who is supposed to study ;D I know nothing about video ai training, even my prompting is still like I would do it with SDXL. It's always about how to get "control". I see some people doing great stuff, I couldn't do. For some amateurs like me, it lacks tools for Area selection + motion control, camera movement, inpainting.... I need to wait until some clever people made this stuff :)

Yup understood - if you're just starting to get into the Comfy world in general, I offer 1:1 LIVE training sessions if there's interest - www.grockster.com/services

That's an interesting theory! I wonder how compositing and blending would look so that it didn't look very layer like?

@GrocksterRox I noticed in a video to video demo, turning an old 4:3 into letterbox, using infill, it tended to work best, during the panning shots. I figure if you at least give the AI a clue what it's supposed to be panning towards, It could save a lot of attention, and therefore work better. At least that's my hypothesis.

Very interesting observation, thanks!

Definitely will take time to get perfect, but it's amazing how far it's come even in the last couple of months.

@@GrocksterRox The same quality as it were even 6 month ago. The same fusing limbs, inconsistency, hallucinations, color bleeding etc.

@procrastonationforever5521 little progress but hopefully will improve

100% right on! :)

Interesting - are you using dpm2 or dpm2_ancestral for your sampler? Also, what scheduler are you using in combination with it? Thanks so much for your thoughts!

@GrocksterRox ancestral give me bad output even on euler.... I use normal scheduler and dpm2

@@robertaopd2182 It's odd, with this workflow I tried dpm2/normal versus the standard euler_ancestral/simple and the latter is giving significantly better output. Maybe try this workflow out with both and post the results. I'm trying to get the optimal output (with consistency) as possible

Dpm2 with the beta scheduler ftw 🙌

@@davidcache6802 jep dpm2 with only 10 steps make almost best like euler.a with 30 steps... for me go better . and i use 3080 10gb.

I'm running a 4090 (24gb vram) but others with 16gb have been able to run this as well (haven't tried less yet)

Yup I totally get that and that's why I'm engaging with the community for this group challenge to see if we can get it better. Huanyuan is very interesting but it's still a TON of time for minimal results (even at a higher quality level), so until it can get even slightly near the same level of speed, it's not going to be a viable option for media creators (unless they're willing to wait months for a significant length clip to come out)

@@GrocksterRox Im looking forward to test your workflow soon. I created a LTX workflow, but as soon as I get "motion" in a video it disintegrates after two or three seconds :( LTX couldn't even generate something "easy" like firerworks for me.... I tried this for some hours. Right now I'm trieing to create a 3 minute decent video with hunyuan... With scenes of 5-18 seconds lengh, sometimes with frame interpolation.... And you are right. Took me already 2 days and it's 3/4 finished now, around 5 hours of pure "rendering" time in total. Media creators could use A100 Gpus and get more out of it with much time. I don't think the consumer hardware right now will be good enough for more. but we are getting there, with a 5090 coming soon :D

Yeah I know it'll get better but with it taking SOOO long for Huan to render stuff, it just isn't a tenable solution for me (and Cog was worse!) That's why I'm trying to get us all into looking at clever/crafty ways to get LTX to produce faster results

@@GrocksterRoxYes. Time is a thing. I remember in the 80's, we had to wait 12 hours or more for our Amiga to render just one raytraycing image that we modeled for hours :D Doing just a 3 minute video, rendering 200 video clips with Hunyuan, remembers me a lot. it sucks :D

@@RagonTheHitman Haha the good ol' days! :)

Great, I'd love to get your thoughts as well as collaboration with the community challenge. Good luck!