I haven't tried InvokeAI, but on the note of alternative implementations, I am a huge fan of ComfyUI. I was only playing around with SD for a little bit because my machine is definitely not built...
I haven't tried InvokeAI, but on the note of alternative implementations, I am a huge fan of ComfyUI. I was only playing around with SD for a little bit because my machine is definitely not built for that kind of workload, so setting up an automated pipeline with Comfy that ran overnights was a godsend. I also really like the visual aspect of it, you can see exactly what steps run into others, and you can switch them around to your hearts content.
InvokeAI has a node based backend (and node editor UI) in our upcoming 3.0 release. Node-based workflow is really nice! ComfyUI looks great. There’s another one called “ainodes”, it’s a desktop QT UI.
InvokeAI has a node based backend (and node editor UI) in our upcoming 3.0 release. Node-based workflow is really nice!
ComfyUI looks great. There’s another one called “ainodes”, it’s a desktop QT UI.
yes! I use Stable Diffusion on AI Horde, it is great, the hands are still messed up but there are some really good models, the quality is definitely not as good as midjourney but that's not the...
yes! I use Stable Diffusion on AI Horde, it is great, the hands are still messed up but there are some really good models, the quality is definitely not as good as midjourney but that's not the advantage of Stable Diffusion, it's the ability to train your own models, LoRAs, TIs etc. and that it's free and open source, also there are a few TIs and LoRAs that fix the hands (EasyNegative, bad hands), as for models, there are many models that are great, Deliberate is an example, it requires more prompt engineering but the results are great and it follows the prompt better than other models, there are also some models that are good even with short prompts, like MeinaMix for anime, I'm unfortunately not very creative, so haven't been able to make any "art" :(
if you would like to use Stable Diffusion but don't have a good enough GPU I'd suggest looking into AI Horde or Colab/Sagemaker studiolab.
I've messed around with it but I've never had an actual need to create artwork so I've never used it in an actual productive manner. I loved the outputs of things like control net though, and I...
I've messed around with it but I've never had an actual need to create artwork so I've never used it in an actual productive manner.
I loved the outputs of things like control net though, and I feel like I would never use anything except for stable diffusion unless that alternative had some form of control net
Midjourney uses stable diffusion but has a lot of secret sauce. You get far less control over the output with midjourney, but it’s also harder to create abominations than with standard SD....
Midjourney uses stable diffusion but has a lot of secret sauce. You get far less control over the output with midjourney, but it’s also harder to create abominations than with standard SD.
Specifically, the feature called ControlNet is a total game changer as far as creative control goes. Midjourney doesn’t support this.
Oh that’s interesting because I struggle with midjourney when I try to create really specific images (I make commercial use graphics) so maybe SD could be helpful
Oh that’s interesting because I struggle with midjourney when I try to create really specific images (I make commercial use graphics) so maybe SD could be helpful
I use on the site Nightcafe, they have daily challenges for people to create and match the theme which has been a good way to learn more about AI art creation. They have different options so you...
I use on the site Nightcafe, they have daily challenges for people to create and match the theme which has been a good way to learn more about AI art creation. They have different options so you can use Stable or Dali 2.
I signed up for MidJourney recently to see the difference and the quality of the end products seem much better.
I do have stablediffusion installed and like to occasionally generate art with it, but my results are less than desirable and quite frankly I’m confused most of the time. Some of the results I see...
I do have stablediffusion installed and like to occasionally generate art with it, but my results are less than desirable and quite frankly I’m confused most of the time. Some of the results I see from people are mind blowing, realistic works of art, beautiful art styles, etc.
Meanwhile mine look fuzzy and like someone went overboard with a white marker for highlights.
It's an iterative process. Many of those fantastic images have been tweaked and edited to get like that. It's never just enter a prompt and get a great result. Try prompts, run batches where lucky...
It's an iterative process. Many of those fantastic images have been tweaked and edited to get like that. It's never just enter a prompt and get a great result. Try prompts, run batches where lucky results (and some weird results), then inpaint hands and faces, pull it into photoshop to clone out bits and bobs, etc.
I've been using Mochi Diffusion, which is a mac native app that supports CoreML, which means that if you have a mac with an M1 or M2 inside it can use hardware built for machine learning. It's...
I've been using Mochi Diffusion, which is a mac native app that supports CoreML, which means that if you have a mac with an M1 or M2 inside it can use hardware built for machine learning. It's pretty fast and extremely power efficient that way #^-^#
I'm a big fan of the InvokeAI implementation of SD. It's got the best UI out there and runs great on my M1 Max MBP.
Thanks for the kind words!
(principal UI contributor for InvokeAI here)
I haven't tried InvokeAI, but on the note of alternative implementations, I am a huge fan of ComfyUI. I was only playing around with SD for a little bit because my machine is definitely not built for that kind of workload, so setting up an automated pipeline with Comfy that ran overnights was a godsend. I also really like the visual aspect of it, you can see exactly what steps run into others, and you can switch them around to your hearts content.
InvokeAI has a node based backend (and node editor UI) in our upcoming 3.0 release. Node-based workflow is really nice!
ComfyUI looks great. There’s another one called “ainodes”, it’s a desktop QT UI.
yes! I use Stable Diffusion on AI Horde, it is great, the hands are still messed up but there are some really good models, the quality is definitely not as good as midjourney but that's not the advantage of Stable Diffusion, it's the ability to train your own models, LoRAs, TIs etc. and that it's free and open source, also there are a few TIs and LoRAs that fix the hands (EasyNegative, bad hands), as for models, there are many models that are great, Deliberate is an example, it requires more prompt engineering but the results are great and it follows the prompt better than other models, there are also some models that are good even with short prompts, like MeinaMix for anime, I'm unfortunately not very creative, so haven't been able to make any "art" :(
if you would like to use Stable Diffusion but don't have a good enough GPU I'd suggest looking into AI Horde or Colab/Sagemaker studiolab.
I followed this and tried but was not good at it. Definitely would want to keep my attention here.
I've tried it before but these days I just use MidJourney.
I've been using it for a few months now. Even with my 2GB GPU, I can create. It's just slow as shit compared to what many people can do.
Yes, it takes a bit of practice with the right training models. But you can make some really good art with it.
I'm getting really into it. Having a lot of fun with image generation, but I'm been playing around a little more with trying to alter video as well.
Oh yes!!
A lot actually, what would you like to discuss?
I've messed around with it but I've never had an actual need to create artwork so I've never used it in an actual productive manner.
I loved the outputs of things like control net though, and I feel like I would never use anything except for stable diffusion unless that alternative had some form of control net
I’m on midjourney and I use it for my professional work, but I’d love to try stable diffusion. Anyone here tried both and care to compare?
Midjourney uses stable diffusion but has a lot of secret sauce. You get far less control over the output with midjourney, but it’s also harder to create abominations than with standard SD.
Specifically, the feature called ControlNet is a total game changer as far as creative control goes. Midjourney doesn’t support this.
Oh that’s interesting because I struggle with midjourney when I try to create really specific images (I make commercial use graphics) so maybe SD could be helpful
Look up ControlNet - that’s the current best technology for controlling image output. It’s magical.
I use on the site Nightcafe, they have daily challenges for people to create and match the theme which has been a good way to learn more about AI art creation. They have different options so you can use Stable or Dali 2.
I signed up for MidJourney recently to see the difference and the quality of the end products seem much better.
I do have stablediffusion installed and like to occasionally generate art with it, but my results are less than desirable and quite frankly I’m confused most of the time. Some of the results I see from people are mind blowing, realistic works of art, beautiful art styles, etc.
Meanwhile mine look fuzzy and like someone went overboard with a white marker for highlights.
It's an iterative process. Many of those fantastic images have been tweaked and edited to get like that. It's never just enter a prompt and get a great result. Try prompts, run batches where lucky results (and some weird results), then inpaint hands and faces, pull it into photoshop to clone out bits and bobs, etc.
I've been using Mochi Diffusion, which is a mac native app that supports CoreML, which means that if you have a mac with an M1 or M2 inside it can use hardware built for machine learning. It's pretty fast and extremely power efficient that way #^-^#