ComfyUI AI: IP adapter new nodes, create complex sceneries using Perturbed Attention Guidance
HTML-код
- Опубликовано: 17 май 2024
- In this video, I show a workflow for creating cool realistic sceneries using the new IP adapter nodes and Perturbed Attention Guidance. This creative test series was amazing, Perturbed Attention Guidance is a great tool together with the new Ip adapter nodes.
Развлечения
Großartig - Danke! (sofort abonniert). Hab gerade AI-Musicvideos via SD für mich entdeckt und mehr Kontrolle über die Szenerie (neben mehr Kontrolle über das Licht, was sich durch IC ja gelöst hat) war noch ein fehlendes Puzzle-Teil.
Very good result thanks a lot
You're welcome and thanks for watching.
very good tutorial, got yourself a new subscriber
Thank you very much and I'm delighted. The next video is about the new IP adapter weight nodes and AnimateDiff. It's crazy what you can do with them.
Great !
Thanks!
wow amazing technique
Thanks for that and for watching!
Great tutorial and workflow as always. Thank you so much! Is the voice from Elevenlabs? That's a good voice for story telling.
Thank you for the kind words and for watching the videos. I'm glad you like them. And yes, this is the voice profile:“ Charlotte”, I just kind of like it. :)
@@Showdonttell-hq1dk Thank you so much for sharing her name. I love to make motivational animation video. I hope you don't mind me using it for my future video :)
thanks for showing us. I sometimes can not choose which output has more quality, cause I see some blur in the generated images and need to sharpen them manually. Is it something common to have blurry images from the first pass or I am doing wrong?
P.S:I use turbo and lightning models while getting these blurry images
Yes, with the Turbo and Lightning models I often encounter the problem with blurred images, especially on the first pass. Although these models are impressively fast, they also have their weaknesses. Especially with very complex workflows. That's why I was so pleased with the Perturbed Attenion Guidance method, you have to try a lot, but after a few attempts the results are simply impressive. However, for images that are meant for professional use, I'm already used to post-processing them with image editing software. Only sometimes do perfect pictures come out of the camera. However, in my videos I try to create everything using ComfyUI's internal tools wherever possible. Additional pixel upscaler nodes (4xUltrasharp Upscaler, etc.) are usually sufficient for resharpening blurred images. As shown in the video, if the NNlatent upscaler node is integrated into the workflow and several Ksamplers are in use, upscaling is often enough for sharp images. But overall the problem is apparently "normal", as you can also see in my uncut versions of the different videos.
Thanks for watching! :)
@@Showdonttell-hq1dk thanks a lot, I am just a noob and you helped me a lot. Hope someday you show us how you do post-processing, it is also another mystery for me.
Thanks again and till next time I watch uncut versions
Wonderful videos and work! Any way you can improve the recording quality though? seems maybe over compressed.
Thanks a lot! I'll look out for the recording thing next time. Thanks for the hint.
Is there a link to the workflow for anyone to evaluate?
You can find and download the workflow on my website: alienate.de. Just click on the image of the thumbnail and drag it into the ComfyUI interface. Have fun trying it out.
The ai voice sounds like Greta 😂
In fact, it is based on a Swedish-British-Irish accent. :)
Please sent me your discord or contact i need your help for building a custom workflow 😊