InvokeAI 3.1 Release - Workflows, SDXL Unified Canvas, and more...

hey everybody you know what time it is it's time for another release video 3.1 is out we have a lot jam-packed into this release so I'm not going to go into great detail on how to use all of the cool new features that we've added the tips the tricks those will come in a following video and I promise I've got a lot more of those coming but today I want to focus on the key features of 3.1 and make sure that you know about all the cool new things coming with the release so let's dig in the first major feature of 3.1 is workflows we mentioned in the last video that a node editor was coming what we didn't share was our full plans for workflows a new capability and a core function of the invoke AI platform going forward I'll go ahead and get the gallery out of the way real quick so you can take in the nodes in all their Glory but if You' use the experimental version of this release you'll notice that the 3.1 version is a significant Improvement we've talked with countless professionals who are using this technology in their day-to-day workflows and for the generation of most production assets a very deliberate and sometimes complicated workflow is ultimately necessary in order to generate the content that's needed with invoke ai's workflow Builder you now have a way to capture this workflow in an easily reusable way so that you can generate not just this image but the next thousand images exactly the same way across your team but what we've also learned talking to professionals who using this technology is that not everybody really wants to use a complicated graph building solution like this and that is what workflows solves for as I mentioned I won't go into the deep dive of building out this graph entirely on this video we'll do that on a follow on but a couple of features that I want to highlight because they're incredibly powerful ways that that the community can create and share these graphs in an easily usable way is the workflow linear view when you're building out a graph that you intend to share with others or that you want to reuse easily yourself you can rightclick the fields that you intend to most often change and make those easily accessible in a simple linear UI so let's go ahead and add our positive prompt and our negative prompt to our linear UI and and in this specific generation I may be okay with the model being the same um let's say I have a model that I'm using at my studio but what I do want to change because this is a seamless graph is I want to make sure that the seamless uh y AIS and seamless X AIS are available and toggleable on the workflow now when I save this workflow and share it with others or if I embed my workflow inside the latence to image node which saves the final image this workflow can be reused and the most important things that need to change to reuse that workflow efficiently can be done using this UI let's go ahead and generate to see what we get you'll notice that as the image was generating the graph execution was visualized the status of each graph can be seen in the top right corner and if you ever want to see the output in any individual node you can click on that node to see its outputs if it's uh text or you know in this case conditioning it'll show the entire object that's being passed if it's an image you'll be able to see that image preview there I've also added a current image node to my workflow so that I can preview whatever I've got selected right now uh so this is our surreal curvy Rainbow Magic Sparkle unicorn pattern that we uh obviously needed uh for the project we're working on when you're saving a workflow you can add some additional details to the workflow including the name a version your author name and a contact information whether that's a Discord or GitHub handle any tags you'd like to be associated with the workflow a short description and notes all this information will be included in the workflow as it's shared out now in this case with a seamless workflow uh let's say for example I'm a game Studio that's trying to create textures for my game games and I want to have a very easy uh reusable texture generator what I'd do is I'd probably include some amount of redefined texture prompt which I'll go ahead and paste in here and then maybe some sort of place folder once done with that we would download our workflow Json and share that file out or as always if we want to embed that workflow the image itself can pass that metadata on and be loaded directly into invoke workflow system let's go ahead and generate some green grass and for the sake of showing you the process of adding a new node you'll notice that my noise node here uh has a fixed seed uh I'll press the space bar to open up our node search window type in a random integer and we'll go ahead and add that in clean this up a little bit and if you ever want to select multiple nodes at the same time you can shift and select and then move all of those together well uh let's go and see what we get now we have our grass image which we can use use or needed or send this workflow to someone else to have them use it the workflow builder in and of itself is worthy of its own release but we've been hard at work we've got even more so let's look at the unified canvas our unified canvas has had some significant upgrades since 3.0 our inpainting process was one of the most complicated nodes and we've done a significant amount of refactoring to make this not only just a clean piece of code but also more powerful than it's ever been let's take a look at the details we'll go ahead and do some in painting but I'm also going to use this opportunity to call out and use every new feature that we've added into the canvas so it's going to take us a little bit of time first things first let's open up our control net settings you'll notice that the control net is now available on the canvas but we've got a couple of options here that make it easy to use uh one of which is going to be this import image from canvas what that's going to do is it's going to take your image from the canvas and Pop That in as a control nut input now the trick here is that you can also use it to save the Box region only so if for example we're doing some incredibly fine-tune details on a closeup here we can actually have that small snip brought in as a control nut so I'll do that now and you'll notice that I've got my kind of canny edge map pulled in from this small snipped image and so now I'll go ahead and brush in my mask what you'll also notice is that all of the settings that are related to ensuring that the inpainted image looks seamless like it was part of the original image have all been brought together underneath compositing settings this includes two steps one of which is the mask adjustment this is how we bring the image in and kind of blur it together and the second is our coherence pass the coherence pass is a second imageo image run that is done as the image is kind of being brought together which is critical for the quality of results that you'll find coming out of invoke new canvas so we'll take one last look at our denoising strength and make sure that it's high enough to get us some different details uh remember we've already got our control nut here so we can get a little bit more free on our denoising strength uh let's even push it up to an eight and we should hopefully find that we still are maintaining our structure right our wizard is looking pretty detailed and quite nice get that demonstration out of the way we'll go ahead and clear the canvas delete out our control nut and we'll switch to our sdxl model with the release of 3.1 sdxl is now capable of using every feature that I just showed you on the canvas we can do text to image Generation image to image in painting control net lauras everything that you've come to expect that you can do can now be done directly on the unified canvas with the sdxl model so I want so I won't walk through this one live and we'll turn this on and we'll do a little mini overdrive here is I use sdxl to generate a new image now that we've covered the major updates to the canvas just a couple of other updates that we've made in 3.1 which will make your life easier as you're using the tool especially if you're a professional looking to use this for your work so we'll go ahead and head back to the text image tab where we'll cover a couple of the other feature enhan ments in 3.1 very often as a professional when you're using these tools you probably find yourself looking to bring in external information into a control net and easily turn that into a generation one of the big friction points there is finding the right dimensions for that which is why in 3.1 when you add an image to your control net you have the option here using the Small ruler icon to set the control image Dimensions to your width and height that'll actually bring those directly up to your width and height now you might want to generate something that's larger or smaller but keep the same aspect ratio and you'll now notice that there's a lock ratio icon that'll keep your width and height the same aspect ratio as you scale up and down so we'll go and lock this and scale it up just a little bit to take advantage of the fact that this model can generate at higher resolutions and we'll go ahead and update our prompt to an elephant on a multicolored rainbow adventure of interactive node-based workflows and we'll go ahead and generate now that elephant is at a party I want to be at for the sake of generating a couple of images here let's go ahead and turn this up to five and generate five more now that we've generated a few we need to go in and pick our favor so we'll take a look at these and that first one that we generated still kind of has a special place in my heart so what I'm going to go ahead and do is hover over the image in my board and hit the star icon that star icon will actually bring that up to the very top and as I generate more pictures it's going to stay there this is a great way as you're developing different concepts within a single Board of marking those that you want to keep and then once you're done with a session or let's be Hest probably 3 months later when you're trying to save some hard drive space you can use the shift key When selecting multiple images rightclick and delete all of those images at the same time as you let out that sweet sweet sigh of relief at having easily managed your Generations remember that these updates are brought to you by a combination of the team at invoke and our open source Community huge shout out to the members of the community who helped us build 3.1 who helped support our users and Discord and gave us feedback and comments as we were building out the workflow Builder we're committed to making this the best way to use this technology as a professional whether you're an indie Creator or a large studio looking to co-create with AI I know I keep saying this but there is even more coming we've got a lot of work to do to realize the vision of invoke AI we're excited for you to use it we hope you'll share your workflows with each other and if you're a developer who's looking to contribute to the open ecosystem of nodes and capabilities of the platform we encourage you to get in touch on Discord we'll see you next time and keep an eye out for new videos to break down nodes control Nets and more coming soon

Channel: Invoke

Views: 11,963

Keywords: stable diffusion, invokeai, ai art

Id: ECbZs5hcD-s

Length: 13min 52sec (832 seconds)

Published: Thu Aug 31 2023