Latent Vision (Matteo) showed me the great power of this model. Especially with the ip-adapter. I hope this wil get embraced by the community. Great video btw. You are one of the few going deeper into this great new model.
Man, i love your videos, and the effort that you put in. Please, please, please make a video on post-processing and raw images. Without altering the raw photograph's subject or background, I only want to post-process it by transferring its style, colour grading, lighting, retouching, etc. from another edited image.
Yes, we have used that with TrioSDXL pack, however not yet with Kolors. Mostly I am used Reactor for Face Swapping, and integrating the workflows. The next Kolors pack will include built in Face model creation and swapping sections on a group toggle.
@@FiveBelowFiveUK, it’s fantastic. Can generate up to 4096 x 4096 images (if you have enough VRAM), so no upscaling and the background details are better as a result.
@@runebinder this is interesting, i'll take a look at that today, the SUPIR upscaling was good, so i wonder if it can take it even further with native HD feeding it. Nice !
I already have SDXL_0.9, SDXL_1.0 and SDXL_fixed, the later being the best one, i have not yet gotten hands on the Kolors VAE, although it is supported by the latest Koda pack. You simply use the new checkpoint and connect it to the Set_vae and Set_model nodes in the backend (V5+)
@@FiveBelowFiveUK Thanks for taking the time to answer. This is a crazy time. I installed Flux yesterday and well.. let's say I didn't have much sleep. Crazy good model. When the dust settles I'm gonna go back to kolors or maybe I'll combine the two. What a time to be alive! ;-)
I hear you, this one is a bit strange. If you see the CivitAI workflow page linked in the description, I have written it up there (click show more and scroll down) to help make it clearer. I also ran through it again in the V3 video, but it's still a bit "non-standard". Usually there would be an all in one checkpoint for comfyui which makes it easier, however it's quite new. Honestly - I spend days downloading these wierd architectures to test them, and usually they do not work after DL 30GB of backend XD or are worse than SDXL base... However - It does mean i occasionally dig something up before people noticed. I'll make a more clear instruction for V4 Pack which is almost ready now !
Latent Vision (Matteo) showed me the great power of this model. Especially with the ip-adapter. I hope this wil get embraced by the community. Great video btw. You are one of the few going deeper into this great new model.
Without Matteo a lot of the stuff we can achieve would not be possible, so we are very thankful for his efforts !
king! i enjoy watching you.
I really appreciate that and thanks for watching !!
Man, i love your videos, and the effort that you put in. Please, please, please make a video on post-processing and raw images. Without altering the raw photograph's subject or background, I only want to post-process it by transferring its style, colour grading, lighting, retouching, etc. from another edited image.
i think i have some idea of what you mean, there is a lot there too, i'll add this to my list. it will likely become a new collection
@@FiveBelowFiveUK awesome thanks a lotttttt
oh btw. you experimented with the face-id yet?
Yes, we have used that with TrioSDXL pack, however not yet with Kolors. Mostly I am used Reactor for Face Swapping, and integrating the workflows. The next Kolors pack will include built in Face model creation and swapping sections on a group toggle.
In the last week I've got this, AuraFlow and UltraPixel. It's been good :)
Enjoy it :) looks like i have to add ultrapixel to my hitlist :)
@@FiveBelowFiveUK, it’s fantastic. Can generate up to 4096 x 4096 images (if you have enough VRAM), so no upscaling and the background details are better as a result.
@@runebinder this is interesting, i'll take a look at that today, the SUPIR upscaling was good, so i wonder if it can take it even further with native HD feeding it. Nice !
sorry. question 3: did you try their own VAE? And if so: did it make a difference with the sdxl vae?
I already have SDXL_0.9, SDXL_1.0 and SDXL_fixed, the later being the best one, i have not yet gotten hands on the Kolors VAE, although it is supported by the latest Koda pack. You simply use the new checkpoint and connect it to the Set_vae and Set_model nodes in the backend (V5+)
@@FiveBelowFiveUK Thanks for taking the time to answer. This is a crazy time. I installed Flux yesterday and well.. let's say I didn't have much sleep. Crazy good model. When the dust settles I'm gonna go back to kolors or maybe I'll combine the two. What a time to be alive! ;-)
thanks man, though all the installing part is a complete mess, didn't manage to follow
I hear you, this one is a bit strange. If you see the CivitAI workflow page linked in the description, I have written it up there (click show more and scroll down) to help make it clearer. I also ran through it again in the V3 video, but it's still a bit "non-standard". Usually there would be an all in one checkpoint for comfyui which makes it easier, however it's quite new.
Honestly - I spend days downloading these wierd architectures to test them, and usually they do not work after DL 30GB of backend XD or are worse than SDXL base...
However - It does mean i occasionally dig something up before people noticed.
I'll make a more clear instruction for V4 Pack which is almost ready now !
🙂