วีดีโอ

Next, I plan to make a small starter module to which you can connect the individual modules in order to build your own small, compact special tool. What do you think, does that make sense? Best regards and thank you for your comment

​@@murphylangathat would be appreciated

Sounds good to me. Not everyone likes very complicated workflows, and there's alot to be said for different workflows or modules for different tasks.

ok, then I will do that soon, I'm currently playing with text to speech F5TTS cool stuff :-)

@colinchambers48 OK, that wasn't the case yesterday. But today I also had some strange behavior from Comfy. Here are the 4 custom nodes that you can look for manually: It is probably this ComfyUI-styles-all github.com/aegis72/comfyui-styles-all SDXL Prompt Styler github.com/twri/sdxl_prompt_styler SDXL Prompt Styler (customized version by wolfden) github.com/wolfden/ComfyUi_PromptStylers JPS Custom Nodes for ComfyUI github.com/JPS-GER/ComfyUI_JPS-Nodes But the workflow should work without it, then only this styler should not work

Were you able to install ComfyUI on your device? It works a little differently than on Windows. But I don't know much about that. The workflow shouldn't have any restrictions if ComfyUI is working. Many error messages are very general. If you see "red" nodes, then not all custom nodes are installed. I haven't tested it at the moment, but it should all work automatically. I'll test it with a freshly installed ComfyUI and publish a list of all custom nodes. Then you can install them manually. The Flux variants and the Kolors variant require more memory. But they aren't switched on after loading, so it shouldn't make a difference. I don't know how that works on a Mac. But if you don't change the workflow for now, it should work. The next reason could be that you don't have all the models that are required. You have to check the "Loader". Maybe select it again.

@@murphylangaThere is no CUDA on MAC. Nevertheless the GGUF Models work fine. You should check if torch uses MPS support.

Hello. I'm sorry, but I don't have a Mac and I have absolutely no idea what you can and can't do with a Mac. It's true that CUDA only works with Nvidia graphics cards, which aren't available in Macs. I don't have anything else to say on this topic.

Thanks for the great comment, that motivates me to publish more in this direction again.🥰

@@murphylanga 👍😁Was an Pleasure help You,Buddy.👍😁👍😁👍😁🇪🇨💙🤟👽

@ВячеславШамрай-й4э 538 / 5.000 You're right, I have a simple solution for you. Right-click on the sampler node and select "reload node" quite far down. In August, some things were changed at Comfy, but unfortunately that meant that some things didn't work 100% smoothly. I didn't completely rebuild the workflow, I just changed an older one of mine. The reload node can sometimes lead to connections being lost. But that shouldn't happen here. Thanks for the info, and have fun with the workflow.

@@murphylanga Thanks for the prompt response, everything works now

Thanks for the tip, I'll take a look at it.

You'll get used to it pretty quickly I think. What I noticed is that I miss the Lora trigger near the SDXL Loras or/and Flux Loras. And I'll have to look into saving for something better. If you have something that you find inconvenient. Don't be shy, just write to me

​@murphylanga thanks. The only thing I've stated to add is big labels above the different groups with the label node. I find it easier to navigate when zoomed out. I always forget which bookmark is which 😅

Name and the bookmark number, why not

Sorry, das kann ich leider nicht mehr machen. Ich habe nur noch das fertige Videofile mit Sprache und Ton als eine Spur. Ich hoffe, in den späteren Videos ist es besser. Damals hatte ich überhaupt noch keine Ahnung, wie man so ein Video für TH-cam erstellt. Danke für deinen Kommentar.

@ChenHaoYan-z1c , "Oh, thank you, you flatter me! In the past few weeks, I haven't even turned on my computer, even though I bought a new one. It only has a 4080, but that's still a huge upgrade for me. Currently, I'm expanding the workflow with Kolors and other features. One thing I know for sure, 96 GB of RAM (I was able to install 2x32 in addition to my existing RAM) makes a massive difference, but 64 GB would be just as good. However, 32 GB is definitely not enough, especially if you want to have fun with Flux and Supir, where more RAM really helps. The 4090 was too expensive for me, but I wish you a lot of fun with it. Thankfully, the weather has cooled down, allowing me to use my computer longer with a RAM upgrade."

@@murphylanga Congratulations on your upgrade! Upgrading to 64 GB is a wise decision, especially with the ever-growing demands of AI content. This enhancement has notably increased the efficiency of using Supir within the entire workflow, enabling it to process tasks in just minutes. The improved performance not only saves valuable time but also boosts productivity significantly. Upgrading the computer was certainly the right choice, enhancing the overall experience and making the workflow smoother and more effective. I am truly appreciative of this improvement, as it has made a remarkable difference.

Well, not really, once you understand the principle, it's easy. But with the heat we have here, it's heading in that direction 🥵 I haven't used the computer for a while and discovered that the last version contained a few errors. Nothing dramatic, but annoying nonetheless. The styler groups were not taken into account as planned. Once the custom nodes and the required models are all installed, it's very easy to use. I'll upload a corrected version later.

civitai.com/models/641359?modelVersionId=795110

@@jenesice No, you don't

@@Dreddoom I only installed the base, no Loras and ControlNet yet.

Looks like there's a big change in the comfyui for flux again 😢 th-cam.com/video/BINp9dxeyU8/w-d-xo.htmlsi=1iS-fwAG4l5VIpzV

Hey, I had seen that too, the guy is really good, but a bit exhausting from time to time. Yes, first let some water flow down the Main, as we say ..

Hi, I looked at it yesterday too, but was hesitant about whether I should install it, it's not a big deal, then I read that it doesn't work with Loras etc. But after your message, it has to be included ;-) By the way, I upgraded my RAM. In addition to the 2x16, I added another 2x 32. And uhh, loading the large flux models is much better. (It was probably moved to the hard drive). I wanted to go to 64 GB anyway because of my CAD job. And I wasn't sure if I could really keep the "old" ones. Now it's 96 :-) That's fine too. I have to get back to work. See you soon.

@@murphylanga Nice, I have 32gb but I used all my slots. I didn't think standard memory would help. Its good to know for the future. Mine is a gaming rig first, so it doesn't make much difference. Don't work too hard 😉

@Dreddoom Hi, I even had time to extend the workflow to the GGUF and made a short video about it. 32GB should definitely be enough. Goodnight, I'm off to bed. Or should I watch Black Myth: Wukong after all, which came with the computer and should be available now...

@@murphylanga Its always time to play...Unless my wife says its not! 😋 Nice work

Thank you very much. Yes, I have not integrated Controlnet and Loras for Flux yet. For now I will use Flux to create a base image and adapt it with SDXL via IMG2IMG or vice versa. There is still too much in flux here, the nf4 was working without any problems recently, but not anymore. I'll just have to observe something first. Matteo has also published a good video, I am excited to see when he adapts his IP adapter to it.

@@murphylanga Yes I saw the Matteo video, it was interesting. I saw this new model today, Ill give it a go. It keeps changing so fast :) th-cam.com/video/nncY3dJLV78/w-d-xo.htmlsi=l2xAe7LqlBer2OV2

@@murphylanga I saw the Matteo video, it was very interesting. I saw this new model today. I just tried it. It seems quite fast. th-cam.com/video/nncY3dJLV78/w-d-xo.htmlsi=F9ODTQj_0vbyLSDK

Hello, @Dreddoom here near Würzburg, it's still sweltering hot. I haven't turned the computer on for a while, just to do "real" work. I had the civitai.com/models/641359?modelVersionId=795110 version installed, but I messed up a few things with the Styler connection. However, I don't see any speed advantage, and the quality is worse. I would rather use the checkpoint that is stored with the Comfy models. With the 64+ GB, Flux works really well. 1.2 sec / iteration. Faster than the other Flux variants and Flux dev is really superior to all other variants. When it finally gets cooler here, I want to look at prompting with large language models. Similar to the blip, but more elegant, and kolors is still on the waiting list. If you want to use the workflow, download the new one on CivitAI, the errors should be gone there. civitai.com/models/641359?modelVersionId=795110

"I'm glad you like it! Don't worry, the video won't be deleted, so you can watch it at your leisure and explore all the exciting features at your own pace, when you have time. 😉Have fun!"

@Dreddoom "I truly appreciate your kind words. Indeed, I've been contemplating a change in voice, still using an AI, but this would require a payment on my part. Luckily, I have a few more months of free access to the "government voice," and with my recent upgrade to a 4080, I'm able to revisit my earlier experiments in local AI voice generation. From what I remember, the results were impressive, though the process was rather slow with only 8 GB of VRAM. Of course, any payment involved would be solely for the AI service, not for the viewers."

​@@murphylangaI wouldn't worry about comments like that. The voice is fine. You can't please everyone, so don't try. In our trade we always say. The job would be great, if we didn't have to have customers 😊. Keep up the great work us idiots cant do

🤗 Thank you very much! I must admit, I'm quite clueless about all this TH-cam stuff.🤣 I just discovered that you can receive notifications when someone replies to your comment. 🤭 I also came across the feature of approved users, moderators, and other roles. Would you mind if I added you to that list?"

Sure, but warn you I'm a terrible introvert. I don't use social media much. So I'm as clueless as you 😊

😂

"The AI voice is quite unappealing. Although I would love to watch your videos, the robotic tone of the AI voice significantly detracts from the overall experience, making it feel like a government training video. I empathize with your criticism, but please understand that these videos are not monetized, and I create them mainly for personal satisfaction. Consequently, I opt for a cost-free AI solution. Granted, I could use my own voice, but this would demand more time and effort, and my English pronunciation is arguably even less pleasant, albeit not robotic. Nonetheless, it is essential to recognize that for a significant portion of non-native English speakers, the AI voice provides a substantial advantage due to its seamless integration with automatic translation services."

@@murphylanga I am only giving you my honest feedback and I apologize if it sounded rude, it wasn’t meant to. I understand your position and maybe you can consider other options, perhaps my opinion is not alone but others simply don’t want to put forth the effort to let you know and just move on. Being an AI expert I’m sure there many alternatives you can explore without a cost involved besides some time. I will keep checking back and continue to subscribe to your channel.

@SouthbayCreations constructive criticism is always welcome and appreciated. Thanks for taking the time to share your thoughts!"

To be fair, it sounds more like an airline flight safety video than a government training video. But I still think you're overexaggerating.

@@blinkonceonsunday1325 😂

Thanks for your comment and the information. I definitely have to test the Kolors model. But I want to wait and see what the new frontend brings and what problems there are. I'm a bit worried about the Rgthree nodes. Have fun testing it, it might not work after an update on August 15th. But I wanted to observe it for a few days before making any major adjustments.

I'll try not to update till I know there's no problems. Kolors was tricky to set up. Followed the guy in the like I sent you. Can't remember his name. He was experimenting use Kolors first and upscaling with flux. He does some interesting things 😊

th-cam.com/video/kVmJv3dmH3k/w-d-xo.htmlsi=5VmDaPMqN2-pIfGC

Thanks for your info, yes this guy is really good, I already subscribed to his channel after your last tip.

@fredougarte9360 The workflow with SDXL definitely works, I use it with a laptop with an RTX3070. The Flux checkpoint models should still work, probably very slowly. But I can't say for sure. Your RAM is important with Flux. Less than 32 GB is probably a problem. But with this workflow it is very easy to get excellent images with SDXL. It should work for you without any problems.

@@murphylanga I have the system horsepower(4090, 128GB), but have been putting off comfy because of the complexity. I'm running Fooocus right now, and all the SDXL loRAs will work with comfy. Just add the drive path. The plan now is to attempt this transcript of yours, and see how it goes.

@lesbec2faces Have fun with it, the timing you've chosen is a bit unfortunate. Comfy is releasing a new frontend on August 15th. It's possible that some things won't work for a while. Either wait a bit longer or install everything first. Maybe start with the latest version, it should all work via the manager. And then don't update any more for a while, I'll check everything and let you know when it works with the new frontend or I've adapted the workflow to it.

@@murphylanga I've had it "installed" for a while. I decided to go back through Comfy Academy by Olivio Sarikas. To hopefully absorb the "workflow" things. In Foocus, I love switching out engines and LoRAs. So, I'm trying to get that "Fooocus" functionality into my workflow. This is all for play for me, so nothing critical at all. Thanks for the heads up about the updates. Eergh.

@lesbec2faces Then have fun, Olivio always does good things. But I only skimmed this series briefly, it's a good introduction. To get a small impression of what is possible, you can simply use the workflow from the video. You should be able to install all custom nodes via the manager (Olivio explained how this works, or I did in my earlier videos). Pull them in and press the queue prompt button. :-)

@allantassie5354 Hello, wow thank you for the wonderful comment. I was very happy. Your idea with the video is fantastic, I will consider it. The computer had a summer break, and today I switched on Comfy on my new machine for the first time. A lot has happened in the 3 weeks, and I first have to sort out what I'm going to put in the next video. A viewer pointed out an interesting model to me and Flux also came out, which is very memory hungry but probably damn good. I'll probably present a first adapted workflow and then see, when the days get cooler, I'll be more in the mood to start the computer again :-). Thank you again for your comment. 💕

Oh no, not for your results, just use the default workflow that comes with Comfy 😉 The effort is for me so that I don't have to constantly think about how to set up a workflow or which of the many small workflows I need at the moment. Here I can switch everything on and off as I please. And of course for the viewers who also want to play with it and maybe copy some modules into their workflows. The videos are something like a guide and a memory aid for me, because I'm very forgetful, so I just watch a video and know how it works again. 😊 The nice comments are then an incentive to create more videos for the viewers who like it and who might want to learn one or two little things..😎

@ragnarlodbroktv9628 - Hello, I don't know this, I had a quick look at what's available. It seems interesting. I think I could simply swap two modules to make it possible to use it. However, I can't promise anything. I'm currently on vacation and don't have time to create workflows. I'll have a closer look at it from mid-August.

@@murphylanga well very interesting to see such a workflow, this colors model is amazing not like other models

and most importantly, this model understands promt very well.

@ragnarlodbroktv9628 Sorry, I haven't found the time yet. Then Flux appeared and on August 15th there will be a big frontend change at Comfy. It will probably take a little longer until I get to it.

@WallyMahar Hello, I'm glad you like it. What do you mean by men and women? Should a male AI be involved, or should more images of men be created? So far, less than 1% of women watch these videos, do you have any tips on what I should change? At least it has aroused the interest of a manager from the fashion industry.😉

That's a very good question and I can't answer it seriously. I don't use Controlnet that much because it slows down image generation on my laptop. I probably wouldn't switch again, especially if I get what I want to achieve. Small info: There is another newer all in one model from Xinsir; diffusion_pytorch_model_promax.safetensors which can also be loaded via the manager.

@@murphylanga I agree with you because the image2image feature has already done a decent job, which sticks to the image by adjusting the denoise value. Despite being time-consuming, a myriad of exquisite models in Controlnet is still very exploratory. Thank you for your reminder; I will check the new model later.

That's a good idea, I pinned your comment. It starts at 30:52

@caphalor1299 The error shouldn't have anything to do with the control net model, I tested the new promax model, it works with the workflow without any problems. Your error message suggests that there is something wrong in the conditioning group (the one in front of the control nets) in one of the 2 red nodes at the very bottom (SDXLneg or SDXL Ref neg), or that something isn't getting there, probably opt clip_wight or opt_clip_hight. Have you entered a value of 1; 2 or 4 in the gray Node Clip Scale in the Start Settings group?

Just try reloading the original workflow and starting it once without Controlnet and then load your Controlnet model there and try again.

@@murphylanga First of all: thank you very much for the quick response and even more important: thank you for sharing this gigantic workflow! I'll try out your tips, but it'll take a while 😉

That would certainly be a possibility, but then you should think about what you really need. I mainly build the workflows for myself for fun and share them with others. An app has no added value for me personally. I still have a lot of ideas that I want to incorporate. This way I am very flexible and can change things spontaneously. But of course the workflow is at a level that would be suitable for an app with a few small adjustments.

@@murphylanga agree this ComfyUI is really endless... thir can reach as far as imagination... insane can't imagine few years ago having this kind of tools

I would never have thought of it without you, thanks again and have fun with it, there is probably a lot more to the Easy Nodes, but Chinese is not my core competence. Um, I mean I have no idea and can't even say hello.

​@murphylanga I couldn't use it without your skills. I'm always watching comfyui stuff. If I spot anything interesting I'll let you know.😊 I remembered where I got the styler from after I saw your link. When I first started using comfyui. It was called shima on gumroad. Very clever modular system. Sadly he had family troubles and stopped working on it. Had fun getting used to your new workflow. Hopefully your channel will grow. Cheers

@Dreddoom It's always cool when you can learn from each other, cheers👌 you or we, I´m not from England😉

@ChenHaoYan-z1c You need to select your checkpoint in the green-yellow Node Checkpoint Loader Preview. Right-click on the name. Select View Checkpoint Info from the menu. If the information about your checkpoint is stored on CivtiAi, a preview image will be displayed there. You can save this as a preview image using use all preview. Or -> Or you can go to the folder where your checkpoints are stored. There you can copy an image with the same name as the checkpoint. Should work with jpg and png files.

@@murphylanga thank you for the response. I just figured it out and clicked one by one a few moments before seeing your comment.

@@murphylanga@murphylanga, and thank you for the sisters. Their integration is also a significant feature that attracts me while seeing.

Manager from Lt Dr. Data, see my comment in your last question. Press Manager, then Custom Nodes Manager. There, under search, enter art or venture. Select comfyui-art-venture and press install. Then restart and that's it. Or manually at: github.com/sipherxyz/comfyui-art-venture

@@murphylanga thx! I have no idea why when i search for it i get no results in the manager, but i just downloaded your workflow and install it with the missing nodes :)

Hello, there are more videos in the playlist th-cam.com/play/PLOlk3DTLbLzQQzMTytsq8oEjdeRhDjI3j.html . In the installation video you will also find more information about the manager. This is the node that everyone should use if they want to work with custom nodes: Here is the link: github.com/ltdrdata/ComfyUI-Manager If you want to save directly under Comfy using the small arrow, you need these nodes: github.com/pythongosssss/ComfyUI-Custom-Scripts Normally, after installing the manager, you can automatically install all nodes, with a few exceptions that are described in the videos. How to do this is also explained in the installation video: th-cam.com/video/-fMgUafa3eI/w-d-xo.html

@@murphylanga thx!

I don't know Zia Fit and I don't use Insta either. I had a quick look to see what it could be. Looks like they use the Face Raactor Node or something like that. There is also Roop under Automatic1111. Both are not released for commercial use! Maybe there is something similar, or better, for which you have to pay license costs. The faceswap model used in the Reactor Node is deliberately "throttled" to prevent abuse. But even if you use it, read the disclaimers and license terms carefully!

@congsonlam5998 Hello, the group with the Reactor nodes is relatively difficult to install. I have published an extra video about this: th-cam.com/video/rR5M7ZjH9HA/w-d-xo.html The installation process should be described in detail there... And also a possibility how you can do everything else except faceswap with the workflow by simply deleting the group without this “complicated” installation method.

@@murphylanga Thank you alot ❤️❤️❤️

@sangheumbyun6749 "You're very welcome! It's wonderful to receive such a kind comment from Korea. I'm thrilled that you found my explanation helpful and appreciate your support by subscribing. Your lovely words have definitely motivated me to create more videos soon. Thank you so much for the positive feedback!"

The download link is under the video description CivitAI and OpenArt

You're welcome, enjoy.