The 2 Ziggys' ComfyUI Adventures: Unveiling Murphy's Upgraded Workflows!
ฝัง
- เผยแพร่เมื่อ 2 มิ.ย. 2024
- Embark on an extraordinary journey into the realm of image generation with the spirited duo, the 2 Ziggys. Join them as they expertly guide you through Murphy's upgraded Modular ComfyUI workflows, incorporating the outstanding SDXL models Mohawk and Cheyenne, masterfully crafted by the brilliant Mr. Aurety. Be entertained and enlightened as these AIs infuse humor and enthusiasm into the tutorial, demonstrating the workflow's impressive new features, including a specialized Corp and Inpaint optimization group and expanded Upscaler groups. Revel in their witty banter as they conjure up stunning images and impart insider tips to maximize your experience with Murphy's enhanced ComfyUI workflows. Prepare yourself for an unforgettable adventure in image generation with the 2 Ziggys at the helm!
Sunday May 19, 2024 12:00
Small fix after inserting the cropped image
Workflow on OpenArt:
openart.ai/workflows/murphyla...
Workflow on CivitAI:
civitai.com/models/461128?mod...
Aurety creator profile on Civit.AI:
civitai.com/user/Aurety
0:00 start
0:08 short workflow overview
0:45 Ziggys briefly introduce themselves
1:43 model Patch Group
2:09 Info setting for pony models
2:38 Kohya Deep Shrink
3:02 Automatic CFG
4:00 Mohawk v2.0 model by Mr. Aurety
4:47 Start First creation
5:53 First result
6:50 Second run with clip set -1
7:13 result 2nd run
7:20 Third run CFG out
07:55 Result 3rd run
08:24 Optimize Group Cropping and Inpainting
15:37 Upscale Groups Settings and Info
18:28 Fascinating end result Mohawk model
18:58 Switch to the Cheyenne model
19:55 First stunning result Cheyenne model
20:13 Workflow optimizations
20:20 Adjustment of Prompt and Lora to the model
21:07 Wow next result
21:27 Switch to Murphy's favorite style - วิทยาศาสตร์และเทคโนโลยี
Very nice work again Murphy. You make it easier to create stunning artwork. Look forward to your next work
😊 So in my upcoming video, I'll be diving into the enigmatic world of Samplers and Schedulers in Image to Image🤯. I must confess, I'm no expert on these topics either, but hey, we'll figure it out together, right? 😉 And don't worry, I'll do my best to keep it fun and entertaining-even if it means throwing in some dad jokes along the way!!
@@murphylanga look forward to it 😉
Really great tutorial and workflow. I'd love to see a tutorial on how to create really great impressionist art that has realistic brush strokes and the appearance of paint thickness. Thus far I've been able to get very promising results low resolution but I haven't been able to figure out how to keep it looking like a real painting as it moves to higher resolutions.
Hello @hatonafox5170 , thank you. Thanks also for your question.
First, try to downgrade Denoise very low in the Ultimate Upscale group. 0.01 is possible and changes virtually nothing to the image. This is a good way to work for paintings.
A second option is to use other upscale models with the Ultimate SD upscalers and the Highresfix. For example, I wouldn't use Ultrasharp, it's good for photos. Try the 4x_NMKD-Siax_200k or 4x_foolhardy_Remacri which can also be downloaded via the manager. There are plenty of others too. Please write to me what you found out. I can also search a little deeper. Best regards and have fun Murphy
Thank you for your suggestions. I will give it a try and let you know how it goes. Looking forward to your next videos and the addition of ControlNet to the workflow!
@@murphylanga So I gave your suggestions a try. I'd say the results were improved but it still doesn't look believable that the images produced were painted. I guess I'd describe the results as being simply a higher resolution version of a slightly blurry or unrefined image. Midjourney actually does a good job at creating oil paintings with impasto effects or clearly defined brush strokes that feel even more real after upscale (most of the time) but I can't get anywhere close to that using stable diffusion. I'm relatively new to this so this might just be a skill issue on my part but I've seen this question a lot and none of the answers I've seen produce realistic results. Any tutorials or workflows you could share would be a big help, I'm sure to many people who like this art style.
I can imagine that what they have in mind is not so easy. Achieving the quality of Midjourney is certainly not easy. Nevertheless, I believe it should be possible somehow. Most people want photorealistic images, so what you and I would like to do is not so common. It might make sense to put a special Lora model in front of an upscaler again. Try to add an artist with your desired painting style to the prompt and possibly enhance the prompt. Maybe an image to image (is like a latent upscale) will also help. There will be a new video tomorrow in which we will go into this in more detail. Or a suitable upscale model. Here is a page where you might find what you are looking for. But I don't really know the site and can't say whether the models there are all “safe”.
openmodeldb.info/
At the moment I want to realize some other things first. But I am also interested in what you have in mind. Best regards and thanks for the information.
@@murphylanga thank you for the additional information and feedback. I’ll give those ideas a try as well. Looking forward to your next video.
I really like your generation of painterly image by text2img. You are awesome.
Is there a way comfyui can convert a realistic portrait photo to a painterly photo without distort facial features of a person? Could you build such a workflow please?
Thank you for your kind words and for appreciating my painterly image generations through text2img! I'm glad you enjoy my work.
Regarding your inquiry, I've already published workflows that can help convert realistic portrait photos into painterly images without distorting facial features. You can find one on CivitAI, and on OpenArt, you can check out the Ultimate V9 Workflow. However, please note that the Ultimate V9 Workflow is quite complex and may be challenging to use.
Additionally, I've also contributed to a workflow on OpenArt that serves a similar purpose. You can find it here: openart.ai/workflows/tamarin_extra-large_98/sdxl-basic-image2image-face-optimiztion/5ki2QaHfXtbOyVFIQjXP
I'm currently working on a new workflow in the Automatic Modular series, which will be easier to use. It's almost complete and is scheduled for release next week. However, please note that it won't include ControlNet support upon release. I appreciate your patience, and I hope you continue to enjoy using these workflows!
@@murphylangaThanks for your reply. I've watched almost all your youtube clips to find out how to convert portrait pictures to painterly pictures without distorting facial feature. I've downloaded V9 workflow but I have even no idea how to start. Your workflows are too complicated and too large to average users. However if you have some time, please try to teach on this topic ( img2imgToPainterlyStyle) and please smaller your workflow if possible. I wonder if comfyui can really convert a people portrait to how much beautiful painting photo.
Hello, next week I will provide something that is very easy to use apart from installing the @suthamchindaudom6410 comfyui-reactor-node. It's not perfect yet, but I think you'll like it. Not the next video, but the one after that ;-) If you were able to open the Ultimate V9 without errors, then you have already solved the problem with installing the comfyui-reactor-node.
For further optimization, I first have to build the modules for Controlnet and Face IP adapter. However, I think that you can have the facial features best with the comfyui-reactor-node. A mixture of everything will then be able to optimize it. But I can't get to that at the moment.
@@murphylanga I appreciate your contributions. When your new workflow finished , please let me know. Thanks
Hello @suthamchindaudom6410 , a new workflow has been published. The faceswap module is still missing to better achieve your goal. But the image to image function is already built in. Best regards