Відео

Kaggle shuts down the notebook when attempting to start up the gradio links and I haven't yet found a way around it. If someone has a way to circumvent this, I can create an updated notebook for it and maybe drop another video.

You can just download and install comfyui locally and upload the workflow. Just be sure to download the models and place them in their respective models folders as well.

My bad. On demand will be a reserved GPU for your session and is non interruptible which should minimize/eliminate premature pod interruptions. Spot however is interruptible and can be paused or stop at almost any moment if supply for that GPU is limited or for any other reason.

I can easily see that becoming a reality one day.

In your activated conda env, run the command: pip install -U rembg

I'm actually working on a few Flux colabs notes that I found for my next video that I think you guys will love. Just give me a bit.

Openhands requires working with docker which can be a headache for most. I'll consider it if there's more requests for it

I have the same! I'm glad to know that it will actually work. :)

In the end, I copied the outputs to my connected workspace folder, then used another template that gave me access to the jupyter notebook. It's an annoyance in Comfy that you can't set an output directory, at least not easily. I did try editing main.py to set the output directory to workspace, but then the restart started to go through the whole startup shebang again, and I couldn't be bothered to wait! You can though save each preview as they come through, but then you have to sit and wait for each iteration.

@@geoffphillips5293 What a pain in the ass. Have you tried using one of runpods Cloud Sync providers and seeing if you can set up a way to push your files there?

You need to download the clip model. You can download it from your comfyui manager. i reference this in the video as well.

@@TheLocalLab yes the problem was you told in video put both model in check point but the one was in need to put in clip and other was in check point. it working with text flawless, but when i do with image it shown me error after sometime :(

​@@amanportfolio Take a screenshot of the error and post it in my discord - discord.gg/5hmB4N4JFc.

Well you won't gain much watching less then a couple minutes of any of my videos. Honestly if you just checked the description, you would have at least possibly spotted the free resources that were available.

I don't believe it watermarks every generation. I think the model was trained on a large amount of stock videos which may lead to watermarks on some videos. I've created videos with the cogvideo's models and was able to generate some without watermarks.

Sorry but this video is a bit outdated. Open-WebUI has made many updates since this dropped that renders some of these solutions useless. I'd advise following up to date instructions from their repo to set this up properly.

Did you only read half the title?

I did. What I'm trying to express is that "running it locally in the cloud" is very incoherent. Locally is a synonym for offline, don't you think?

@@petneb I guess it can have different meanings depending how you look at it. The whole point of this set up(at least for me) was to have the Qwen model connected to projects I have running locally on my device instead of running them on a web servers. I have projects that I run on local ports that every so often connects to the internet for different tasks, does it make it any less local?

Try the LTX video and Hunyuan Video models. I think they might be a bit better in quality.

same here

You could possibly get away with using the 2B model, but I'm not fully sure. You can try and see.

I haven't created a one click installer for this project yet but will consider it if there's a need for it.

I haven't seen any implementation of custom Loras for LTX video yet but if you find anything, let me know and I may cover it here on the channel as that would be a great addition to these LTX workflows.

If your using the basic workflow, it should save to your outputs comfyui folder as a webp file. Also, I have a node in my advance workflow that will save the final output in MP4 or GIF format on my Patreon if interested.

up

@@TheLocalLab Ok, thanks. I have manage to change the file to mp4, thank you for your help. I will check your Patreon.

I solved it installing the custom node 'VideoHelperSuite', then add a new node with 'VideoCombine', set it to h264-mp4 format and 24 fps (fps of the generated video), then connect the 'VideoCombine' node to the 'VAE Decode' image output.

@@noobhash Yeah this should work as well.

My pleasure, have fun with them.

yes but they do not work. mycomfyui cant automatically detect what LTXVConditioning or 2 other nodes are. I think you need to buy his patreon post?

@@monoham1 No you don't have to purchase anything on patreon to make the basic workflows work. However you will need to update ComfyUI to the latest version if you haven't already.

And I appreciate your comments. This video model drop is truly amazing and will probably be the start of some more amazing models to drop next year. Can't wait.

Yeah I've been getting this error with a variety of other workflows as well. I believe this might be a ComfyUI bug.

@@TheLocalLab Thanks for the reply. So this is a work flow error? With a different workflow it should work?

@@64z I believe its a bug in a recent Comfyui update that affects all workflows that utilizes the CPU and GPU during inference. If your on windows, you can try downloading and using an older release version of Comfyui and seeing if the issue persist. Comfyui release page - github.com/comfyanonymous/ComfyUI/releases.

@@TheLocalLab Thanks for the suggestions. Will see if i can do that.

No, Ooba Booga doesn't have a extension for sadtalker.

@@TheLocalLab ok how to make the same process to access sadtalker webui in colab?

Port error for comfy? The default is 8188.

@@TheLocalLab no in open web, it just says it doesnt exist even when copy stright from cmd

@@davisj789 What's the url or port that your pasting?