ComfyUI: CosXL, CosXL Edit InstructPix2Pix (Workflow Tutorial)
HTML-код
- Опубликовано: 8 июн 2024
- This tutorial focuses on CosXL and CosXL Edit InstructPix2Pix workflows for ComfyUI. The workflow tutorial video includes all parameters explained, advanced model merging in comfy UI for converting any model to CosXL, the upscaling technique used with CosXL and CosXL edit, and some tips and tricks to get the best-desired outcome.
------------------------
JSON File (RUclips Membership): / @controlaltai
CosXL Stability AI (Hugging Face): huggingface.co/stabilityai/cosxl
Tile ControlNet by TTPlanet (CivitAI): civitai.com/models/330313/ttp...
4x Foolhardy Remacri Upscaler (Hugging Face): huggingface.co/uwg/upscaler/t...
------------------------
TimeStamps:
0:00 Intro.
00:58 Requirements.
03:48 CosXL Base Workflow.
08:22 Dual vs Single CFG Node.
09:43 CosXL Base Refined.
12:38 Pixel Upscale.
20:25 CosXL Edit Workflow.
27:05 CosXL Edit Parameters, Examples, Upscale.
34:25 Advance Model Merge.
44:27 Additional Tips. - Хобби
A wonderful tutorial! I learned a lot more about ComfyUI. Thank you so much for taking the time to create this. Also, showing how you created the workflow and following along myself works much better as a learning tool.
Thank you!!
One of rare gem channels in SD world. Awesome content. Would be so nice to see your take on fixing hands in comfyUi!
Thank you!!
Agreed 100% and I too would love to see what can be done about fixing hands
Best channel! in ComfyUI! I liked the content, so I signed up for membership.😁👍
Thank you!!
I was about to sleep but now i gotta keep watching for half hour more haha. Keep up the great work.
I am a Beginner in ComfyUI and it is hard for me to understand all. But i love it. Thx for the great work. Learn more and more when i watch your videos.
Absolutely incredible
Thank you!!
Thank you for the fantastic lecture~!~!!😍🤩😍
wow what a great video watched through the whole thing and set it up myself. I also realized i was upscfaling my images wrong (facepalm). Im going to binge your videos great work buddy!!!
Thank you. Btw there is no wrong way to upscale. Every way is the correct way only different....😉
Amazing Tutorial! Why is it better to use the sampler custom advanced instead of a default k-sampler with all the widgets turned to input nodes? For organization purposes?
Thank you!! More advanced and fine tune controlling for CosXL. For best results use sampler custom advance.
wow! AMAZING. One question, can u merge cos edit and realvis? and edit images with the quality of realvis?
and one more question, an u use ipadapter with cosxl edit?
Cosedit edit causes some errora but works. I will try it. Ip adapter I have not try. What are u trying to do with ipadapter let me know I will try that.
@@controlaltai it would be amazing for example to change the material from a sofa and give it the reference of the material with ipadapter. i was trying that.
You don't need IP adapter you can describe the material. Thats The point of cosxl edit. Just say plastic sofa, leather sofa etc and it should work.
@@controlaltai yeah but what i need is the exact texture, imagine a pattern and i need to edit the sofa with it. Thank u to answer
Thanks for this tutorial. I'm having a problem. The cosxl_edit.safetensor model in Pix2Pix is returning a grainy green image that barelly resembles the original image. Any hint for me, please?
Check the dual cfg connections. They are tricky. Recheck cfg. Ensure image scale to total pixels is correctly connected. Close command prompt and restart comfy from cmd. Try again And let me know.
@@controlaltai Thanks for the answer! I checked the connections you mentioned. ComfyUI is updated. I downloaded safetensor again. And lastly, I looked for a similar workflow on civitai. I keep getting the same result. Now I just installed a clean Comfyui. Still the same problem. Really weird. Anyways, thank you for this video. You explain very well!
Send me an email with your workflow, image and the prompt. I will have a look. There must be something wrong, I don’t think it’s comfy. mail @ controlaltai. com (without spaces)
@@controlaltai I sent the email. But it's working... For some reason the 2 first times I downloaded the safetensor, it was corrupted. -.-"
3rd for the charm. Thank you so much! It's working and It's beautiful!!!!
Treeeemendous video like always!
...although too advanced for me 😅😭😅
✨👌😎😮😵😮😎👍✨
What's the minimum VRAM requirement?
I don't think it's anything specific for just the model, it's a simple sdxl based model just fine tuned. Whatever is required for SDXL, same here. Except for the upscale. If you don't refined or upscale 6 to 8 gb would work.
Hi, great work! I get this while merging cosxl_edit model with another sdxl model ->>> WARNING SHAPE MISMATCH diffusion_model.input_blocks.0.0.weight WEIGHT NOT MERGED torch.Size([320, 4, 3, 3]) != torch.Size([320, 8, 3, 3])
Any thoughts?
Are you merging the sdxl model?
Yes finetuned sdxl model. Do u know how many ram u need for saving a merged checkpoint?
The cos xl edit model always will give an error. It's not designed to be merged. However it does work. But I would not recommend it. Vram is not an issue here. Merge only the cos xl model not the edit one.