Great video, I just subscribed:) also can you please tell me the extension you use that show the different prompts and how many items they were trained with (2:49)
thank you, I think you are referring to "a1111-sd-webui-tagcomplete" it is an autocomplete extension , you can find it in the extensions section and install it ... not sure what you meant with training items since this video is about outpainting only, so if it is related to the script, I dont know unfortunately.
this might be easier in some SD versions than others, i suggest you to try using Controlnet or using inpaint method that i have described in the two other videos in ua-cam.com/video/8JY1ApvdAjA/v-deo.html and in ua-cam.com/video/7IjJCEk-2mM/v-deo.html this can be tricky sometimes, and you might neet to repeat the generation and play with the denoising level several times till you get the hangs of it.... in general SD still requires alot of try and error to get things right compared to Photoshop generative fill for instance .... so it is normal if results are not always as expected especially in realistic SD models
true, current features including Controlnet do not offer enough flexibility without relying on external tools for editing, cropping, masking...this can be disappointing, but i expect in the upcoming months, these features will improve, hopefully.
Thank you! You are awesome! Exactly what I was looking for however, I don't have that 'Off' button like you do by your width x Height. How can I get that?
Glad I could help! GUI difference could be due to the Web ui interface version, which continues to change, but the principle remains the same ... off button disappeared yesterday with me with upgrade to A1111 version 1.4
@@AI-HowTo It's better to use inpainting model for 2nd outpaint, or further (works great for background). With first one it tryes to generate girl (even if you have nothing related to it in prompt).
this may happen depending on used SD checkpoint/image ... testing alternative methods such as Controlnet as here ua-cam.com/video/8JY1ApvdAjA/v-deo.html could provide an alternative better method, in essence it is experimental to see which approach work better for a certain use case.
not sure why, you can try other methods, they maybe more effective, depending on the SD being used, using Controlnet method for outpainting for example which i explained ua-cam.com/video/8JY1ApvdAjA/v-deo.html or using inpaint tool explained here ua-cam.com/video/7IjJCEk-2mM/v-deo.html the steps must be followed carefully even if the GUI changed slightly, because A1111 continues to updates its GUI slightly every now and then. ... often poor man's outpaint should work gradually... using img2img with increasing the size of the image and choosing (Fill) opion should also work, so basically there are 4+ methods for outpainting in A1111
This person has no idea what he is doing, showing such dumb method, more harm than good. What you REALLY should do in this case - open pose - create scene with any woman in required clothes, when done, transfer face from portrait to that body. But it is too smart, method for this "expert"...
welcome to your SD channel ! and thanks you
the focus on outpainting technics is an excellent idea ! very interesting tuts
Thank you
Great video, I just subscribed:) also can you please tell me the extension you use that show the different prompts and how many items they were trained with (2:49)
thank you, I think you are referring to "a1111-sd-webui-tagcomplete" it is an autocomplete extension , you can find it in the extensions section and install it ... not sure what you meant with training items since this video is about outpainting only, so if it is related to the script, I dont know unfortunately.
None of these techniques have worked for me, SD just generates a brand new image under the original image with no continuity
this might be easier in some SD versions than others, i suggest you to try using Controlnet or using inpaint method that i have described in the two other videos in ua-cam.com/video/8JY1ApvdAjA/v-deo.html and in ua-cam.com/video/7IjJCEk-2mM/v-deo.html this can be tricky sometimes, and you might neet to repeat the generation and play with the denoising level several times till you get the hangs of it.... in general SD still requires alot of try and error to get things right compared to Photoshop generative fill for instance .... so it is normal if results are not always as expected especially in realistic SD models
@@AI-HowTo I've been using control net but I'm kinda bummed that it doesn't have the option to outpaint in only one direction
true, current features including Controlnet do not offer enough flexibility without relying on external tools for editing, cropping, masking...this can be disappointing, but i expect in the upcoming months, these features will improve, hopefully.
Thank you! You are awesome! Exactly what I was looking for however, I don't have that 'Off' button like you do by your width x Height. How can I get that?
Glad I could help! GUI difference could be due to the Web ui interface version, which continues to change, but the principle remains the same ... off button disappeared yesterday with me with upgrade to A1111 version 1.4
You could try switch model to "RevAnimated Inpainting" for better results. Or lower Denoising Strength to "0,55". Or combine those two approaches)
Thanks
@@AI-HowTo It's better to use inpainting model for 2nd outpaint, or further (works great for background). With first one it tryes to generate girl (even if you have nothing related to it in prompt).
the generated outpaint fades downwards for some reason
this may happen depending on used SD checkpoint/image ... testing alternative methods such as Controlnet as here ua-cam.com/video/8JY1ApvdAjA/v-deo.html could provide an alternative better method, in essence it is experimental to see which approach work better for a certain use case.
perspectives of left and right side of the woman do not match.
Thanks a lot.
Most welcome!
maghribi ?
no
use outpainting mk2 is much better ,it remove the seems completely by agistment mask of outpainting mk2 script
very creamy
Followed exacty, but all it does is creating a whole new/different generation, rather than expanding the current image.
not sure why, you can try other methods, they maybe more effective, depending on the SD being used, using Controlnet method for outpainting for example which i explained ua-cam.com/video/8JY1ApvdAjA/v-deo.html or using inpaint tool explained here ua-cam.com/video/7IjJCEk-2mM/v-deo.html the steps must be followed carefully even if the GUI changed slightly, because A1111 continues to updates its GUI slightly every now and then. ... often poor man's outpaint should work gradually... using img2img with increasing the size of the image and choosing (Fill) opion should also work, so basically there are 4+ methods for outpainting in A1111
This person has no idea what he is doing, showing such dumb method, more harm than good.
What you REALLY should do in this case - open pose - create scene with any woman in required clothes, when done, transfer face from portrait to that body.
But it is too smart, method for this "expert"...
wow, you seriously have issues! you really need help.