Відео

You have to update comfy. then update the custom node. close everything and then restart. Check for fetch updates again to ensure you are on the latest version of the node.

@ thanks I’ll try again

That is a different GitHub repository. Install ControlAltAI nodes from Comfy Manager. github.com/gseth/ControlAltAI-Nodes

Please read description. I have explain how to install xformers in the requirement chapter on video.

This is the initial release. I have to explore integration of ControlNet. I have mentioned all this in the video. The current pipeline was complicate to reach at this level of stability. I have to test and see if ControlNet can be added and if possible or not.

I know that image to image doesn’t work in this pipeline. So I guess after that redux should be done. I haven’t tried redux directly as I was in the middle of this when they released the tools.

All good, I just thought that would have so much potential and if it could be done. Nice video by the way, really nice work. 😊

Thanks. It was very hard to get this thing stable. Will research further what else can be added like ControlNet image to image, etc and how.

You are talking about this? github.com/attashe/ComfyUI-FluxRegionAttention Your code is not exactly used. The pipeline is completely different. I checked your repository some overlapping is there for Xformers attention. But it is beyond that. BBox and Mask feathering are there in Attention. Conditional strength is there. Attention handling is different taking all these things. Makes and BBox both are going in attention. Attention Cleanup is done. We worked on the BBox and Masking for over a month and a half. Unfortunately I cannot Update the Video, But I can update the description with your github link as "Base Implementation from Attashe: github.com/attashe/ComfyUI-FluxRegionAttention". And I can update my github page. Let me know if thats acceptable to you and apologies if there is an oversight on our part. Edit: Please check, I have made a Pin Post, Updated Video Description and included Acknowledgments in The GitHub page.

Thanks, I see that you made a lot of changes and obviously not against copying code parts

All things have changed in comfy. Clip seg is not supported. Ultra lyrics was compromised. It now requires a separate install. A whole new workflow has to be created.

@@controlaltai hey thanks for your reply. is there any similair mechnaic for face enhanement which is working ? 13.50 is what i need... please make a new video for us.

Yes Sam 2 or grounding dino. There is a custom node called layer style. It has multiple segmentation options. Will try and make a new face detailed video.

@@controlaltai Awesome. Looking Farward to :)

SUPIR requires a license. If you don’t have a powerful system you can rent a you and run workflows there. Probably like run pod or something.

Depends on use case. Some stuff which I showed here supir cannot do. But there are some things supir can do. For flux most of the corporates use with some brand text etc or want too much consistency like 95%. Supir cannot handle text. Flux handles consistency very well without adding much details. Choose whatever suits your use case.

@ i see. for preparing dataset of a real person lora, which should i use do you think? My problem is the full and half body images of this person are not that great, so i was thinking to upscale these images

Lora, try supir and see if the skin texture and all are realistic and it’s not adding too much. If satisfied stick to supir, if not then switch to flux. Telling to try supir first since it’s easier and quicker, flux is not that straight forward.

@ thank you so much!!!

Something wrong with the model. Ensure checkpoint is sd1.5 and not flux or sdxl

Flux schnell is way behind mid journey you need flux dev or pro which is far better than mid journey.

@controlaltai both are paid

Not sure, could be the resolution. of the image. Check that and check the folder structure if it is correct as its a bit tricky, multiple models have to go in multiple places.

@@controlaltai the new error I get is about weights? Would that be linked to the image? "up_blocks.1.attentions.2.transformer_blocks.0.norm1.weight, down_blocks.0.attentions.1.transformer_blocks.0.attn2.to_out.0.weight, up_blocks.3.resnets.0.norm1.bias, up_blocks.1.resnets.2.conv2.weight, up_blocks.3.attentions.1.transformer"

Let me do some research and get back to within a day. The image dimension give mis match tensor errors. This is giving transformer attention blocks errors. Can you tell me if you are using legacy comfy or latest version?

@ i am using the portable version. I updated about a month ago. Python 3.10.

@@Nibot2023 ok thanks, let me check and research for you what is this error. Will get back shortly.

Yeah comfy changed everything, and it requires developers of the custom node to update their nodes. They custom nodes from ZhoZhoZho are never updated to be honest. I stopped using them. You can get an alternative yolo from layer style node. Basically the tutorial concepts are valid, yolo world can be replaced with SAM 2 for example. Anything that can do prompt detection would work.

Use the birefnet from layer style custom node. It’s basically the same.

It’s only for paid channel members. You can however watch the tutorial and create the workflow yourself. Everything is shown.

Not yet, Give me some time. I will release the fix, but won't be able to make a full tutorial on it. The basic thing is you can replace media pipe with SAM 2, and there is no need for clip seg. Blip is outdated and Florence can be used for that.

@@controlaltai Sure, pls don't feel stressed on that. Thanks for your response. BTW: do you know how to install the YoloWorld successfully? Throws an error on while installing YoloWorldESAM: "AttributeError: module 'pkgutil' has no attribute 'ImpImporter'. Did you mean: 'zipimporter'"? U have a reliable Link on how to set it up successfully? Thank you for your detailed and explaining work by these tutorials on ComfyUI; I really appreciate that. Regards, Roger

Yeah those are different. Will explain them in a tutorial. However highly recommend you watch this one for the logical flow understanding.

@@controlaltai Yes! I am excited to see what they can do, while grateful for the efforts other groups who have been making canny and depth, inpaint, and ipadapters, I have not been impressed. I have much higher expectations with these. I took a brief look at the Comfyui site and it looks like they have native support and a few examples to play with.

I have done a union pro tutorial. Next video is about flux region. We made custom nodes for that very cool concept. After that video will make a full workflow tutorial using these tools as well. We can do some amazing restoring and conversion using these tools union pro.

I'll be sure to check them both out. Thank you! :)

ComfyUI\models\ipadapter, if there is none create it, or install from the manager any ip adapter model, it will create itself.

Please watch part 1 of the video on how to install ollama. GGUF is already explained in this video.