Going forward, I think this is the right direction for any and all AI tools: either creating plugins or tabs or having Adobe-like pull down menus (scaling, editing, effects, etc). Most of the AI tools require so much work beforehand just to create and adjust the image. As you pointed out, "learning nodes" is time consuming and takes away from the creative process.
from what i understand - the model you choose for your iterations (creativity) will need the correct scheduler selected for it, especially if its a lightning or turbo configured model.
5 місяців тому+1
Great tutorial! I'm sure it's a silly question, but I'm a beginner in using Invoke: where exactly can the 4xNomoswebphoto_ersgan, CustomXLv7 and KDPM2 models shown in the upscale-tutorial video be downloaded and installed into Invoke AI?
Missing the Denoising Strength for Upscaling. I hardly get any sharp images. Imagine a wool pullover where you can see all the stitches, upscaled it looks like a cotton shirt. I turn down Creativity, beyond -8 the server crashes, and even -8 does not do clear, detailed images. I tried other upscale models like "4x-Ultrasharp" - same result. What am I doing wrong?
My machine is taking something like 20 minutes to upscale a 512x512 to 2048x2048. Its only giving me the usual AI Generation Models under the 'Generation tab'. I need 25 steps to make it look decent using Juggernaut. Is that normal?
As a photographer and early adopter of AI image generation (using it in specialized Android apps long before it became mainstream online), I often wonder what use cases most people have for upscaling images. Even for photos or visuals intended for movie poster sizes (60cm x 90cm), t-shirt prints, or typical art prints hung on apartment walls, ultra-high resolutions aren’t necessary. After all, people don’t examine them with a magnifying glass. In photography, high megapixel counts are mostly beneficial for cropping-a scenario that doesn’t typically apply to AI-generated images. In 2022, I worked on an art exhibition project using old "DALL-E Mini" images, which had a native resolution of just 265x256 pixels. Despite their low resolution, they still worked beautifully as abstract art pieces. So, I find myself asking again: what are you doing with AI-generated images that require such enormous resolutions? Most of them are just shared online, where high resolutions aren’t needed. Are you all creating posters to cover the sides of buildings, or something? 😊
Absolutely one of the best tools for creators out there!
Going forward, I think this is the right direction for any and all AI tools: either creating plugins or tabs or having Adobe-like pull down menus (scaling, editing, effects, etc). Most of the AI tools require so much work beforehand just to create and adjust the image. As you pointed out, "learning nodes" is time consuming and takes away from the creative process.
I just hoooope we will get layers for Inpainting soon too!
Take a look at #new-release-discussion in our Discord for a little sneak peek :)
Awesome Thank you for Sharing 💯✴
from what i understand - the model you choose for your iterations (creativity) will need the correct scheduler selected for it, especially if its a lightning or turbo configured model.
Great tutorial! I'm sure it's a silly question, but I'm a beginner in using Invoke: where exactly can the 4xNomoswebphoto_ersgan, CustomXLv7 and KDPM2 models shown in the upscale-tutorial video be downloaded and installed into Invoke AI?
support.invoke.ai/support/solutions/articles/151000195887-upscaling-models-community-edition
@@invokeai Thank you very much!
will you add flux ?
Missing the Denoising Strength for Upscaling. I hardly get any sharp images. Imagine a wool pullover where you can see all the stitches, upscaled it looks like a cotton shirt. I turn down Creativity, beyond -8 the server crashes, and even -8 does not do clear, detailed images. I tried other upscale models like "4x-Ultrasharp" - same result. What am I doing wrong?
If you jump into our Discord channel and share settings, we can help troubleshoot there!
I don't see a slider with is the option of shortcut key for this?
My machine is taking something like 20 minutes to upscale a 512x512 to 2048x2048. Its only giving me the usual AI Generation Models under the 'Generation tab'. I need 25 steps to make it look decent using Juggernaut. Is that normal?
Hard to know without knowing more about your system. Hop into our Discord channel and they can support you there.
As a photographer and early adopter of AI image generation (using it in specialized Android apps long before it became mainstream online), I often wonder what use cases most people have for upscaling images. Even for photos or visuals intended for movie poster sizes (60cm x 90cm), t-shirt prints, or typical art prints hung on apartment walls, ultra-high resolutions aren’t necessary. After all, people don’t examine them with a magnifying glass. In photography, high megapixel counts are mostly beneficial for cropping-a scenario that doesn’t typically apply to AI-generated images.
In 2022, I worked on an art exhibition project using old "DALL-E Mini" images, which had a native resolution of just 265x256 pixels. Despite their low resolution, they still worked beautifully as abstract art pieces.
So, I find myself asking again: what are you doing with AI-generated images that require such enormous resolutions? Most of them are just shared online, where high resolutions aren’t needed. Are you all creating posters to cover the sides of buildings, or something? 😊
You're upscaling things that don't need upscaling though.