Hi Streamy! Nice work. concisely eyplained. I think of training a model on style. (Friendly version of A. Tomne novels). Are models the right approch for thsi purpose, and do you have any hints? Thx andvance
yes training a lora model is the current only approach to good control, multiple models I feel is the best approach to higher details, so close up for the close ups, mid shot trained for waist up, lora model for the background style etc even multiple face so on close up shots you could have a happy face trained to use on those and a sad face etc trained for when needed on those scenes. I see it as breaking everything down like a story board and assigning models to the story board parts to build the scenes wanted. more time consuming to start with but once there down you have a consistent data base of the assets required.
you can type the lora www.youtube.com/@aclysia is the maker of the plugin I have left comments and her reads and relys etc and could implement it if asked or possibly give a reason of issues it causes. it be work mentioning it. You can link krita diffusion to your Comfy but to what I can understand the plugin will only use that add-ins that its made to use, so i am not sure it work with custom work flows, Id imagine it uses its own.
civitAi has a lora trainer costs 500buzz to make a lora, but no khoya headaches. I'd say atm that the easy way. there is also control layer face in krita ai where you just put the face image but its not as good as a lora
Hey man, thanks for your videos. I trained my model via the Replicate website and downloaded it. I followed your steps in this video to impliment it on Krita, but my face is just not showing up. Do you know how to resolve this? Thanks a bunch.
are you using it with a correct model, krita does not put a error anymore i found if i use a sd1.5 lora with sdxl but it won't do anything so check that and try heavier weights on using the lora
that's this video, just using the lora in krita, only you will need to install kohyaSS to do it, it does have great results though. but every has issues getting kohya working, my kohya videos are probably out of date. ruclips.net/video/yB3_pW1Ev0I/видео.htmlsi=tgLpXb2apaZHa-4l
Hi Streamy! Nice work. concisely eyplained. I think of training a model on style. (Friendly version of A. Tomne novels). Are models the right approch for thsi purpose, and do you have any hints?
Thx andvance
yes training a lora model is the current only approach to good control, multiple models I feel is the best approach to higher details, so close up for the close ups, mid shot trained for waist up, lora model for the background style etc even multiple face so on close up shots you could have a happy face trained to use on those and a sad face etc trained for when needed on those scenes.
I see it as breaking everything down like a story board and assigning models to the story board parts to build the scenes wanted. more time consuming to start with but once there down you have a consistent data base of the assets required.
I miss ipadapter in the Krita plugin or the ability to use a custom workflow. In a lot of cases it would eliminate the need for a lora
you can type the lora
www.youtube.com/@aclysia
is the maker of the plugin I have left comments and her reads and relys etc and could implement it if asked or possibly give a reason of issues it causes. it be work mentioning it.
You can link krita diffusion to your Comfy but to what I can understand the plugin will only use that add-ins that its made to use, so i am not sure it work with custom work flows, Id imagine it uses its own.
Phenomenal! Any comfyUI Krita workflows?
Search and find.
right now for me i am using just the krita and its built in work flow, I like simple.
I will probably look at Comfy later down the road,
@@streamtabulous ok nice I’ll start with what you e shown us, thanks 🙏🏽
welcome back
where did you upload your image?
Just getting started on this - are there any easy methods to create my face plugging since you posted the video?
civitAi has a lora trainer costs 500buzz to make a lora, but no khoya headaches. I'd say atm that the easy way. there is also control layer face in krita ai where you just put the face image but its not as good as a lora
Hey man, thanks for your videos. I trained my model via the Replicate website and downloaded it. I followed your steps in this video to impliment it on Krita, but my face is just not showing up. Do you know how to resolve this? Thanks a bunch.
are you using it with a correct model, krita does not put a error anymore i found if i use a sd1.5 lora with sdxl but it won't do anything so check that and try heavier weights on using the lora
love your work but can you make a video on how to train a model like you did with your face? Please make a detailed video if possible! Thank you!
that's this video, just using the lora in krita, only you will need to install kohyaSS to do it, it does have great results though. but every has issues getting kohya working, my kohya videos are probably out of date.
ruclips.net/video/yB3_pW1Ev0I/видео.htmlsi=tgLpXb2apaZHa-4l
ps. i like rendering a image and using a1111 and this method also, gives great results
ruclips.net/video/y4xspq6YEbA/видео.htmlsi=9pqOYxa_y2HCEfn4
how did you make the myface safetensors 2:28 ?
used kohya, intall has changed since my original video to install kohya.
ruclips.net/video/yB3_pW1Ev0I/видео.htmlsi=6njPGV65q2VuIQY-
ruclips.net/video/6T-cmomHWe0/видео.htmlsi=vEEFnteu30dXXyr1
How to make the LoRa of your faces?
ruclips.net/video/yB3_pW1Ev0I/видео.htmlsi=0B5CgQKSalThGR_n