Invoke

I am honestly more excited than is probably reasonable about the conversation that Bennett and I will be having at SXSW - If our past few conversations are any indication, they'll be raw, philosophical, and full of insight. He's a pioneer, and one of our most creative users - What he's done with Invoke is nothing short of incredible, and I am eager to see the response to his work as he begins to share more of it in 2025.

Watch as we take a basic 3D blockout of a tavern and add in some texture and environment details to see what it could look like fully rendered. Super useful for testing out different concepts quickly or for a 2D specialist to give a 3D specialist more guidance on how a final asset should look.

The response to this session has been fun -- Our customers in gaming have begun to regularly use this workflow of control compositing for their work.

Excited to take the GDC Stage at the AI Summit New York to push past the hype of "Generative AI" across Media & Entertainment, and take look at the practical applications we're seeing evolve in Gaming. I plan on talking about the latest in intellectual property, model development/training, and emerging applications - If there's anything else you're interested in hearing, shoot me a DM. See you there! #GDCxTheAISummit

LF AI & Data welcomes the Open Model Initiative (OMI) as our newest Sandbox project! 🚀 Backed by a passionate community and leading organizations like Invoke, Civitai, and Comfy Org, OMI is set to push boundaries in ethical, accessible, and competitively capable AI models. From open licenses to robust governance frameworks, OMI shares our commitment to advancing open AI for all. 🌍 "With the integration of the Open Model Initiative, we are propelling our efforts to make AI universally accessible and advantageous. This initiative allows us to foster an environment where innovation and advancement in AI can flourish freely and without constraints." - Ibrahim Haddad, Ph.D., Executive Director of LF AI & Data 🔗 Learn more and get involved with the OMI here: https://lnkd.in/dBzwusae #opensource #linuxfoundation #oss #lfaidata Kent Keirsey Devon Hopkins

Incredibly proud of what the Invoke team has accomplished in building human-centric tools, and for being featured in Liam Deane's latest research from Omdia as leading the way in 2D asset creation. “Moving towards creation of assets that will actually be placed in game is more demanding. However, a number of companies like Invoke… are focusing on developing effective specialized tools for game artists for both concept art and production assets.” I appreciate our customers who spoke to how Invoke is being used effectively in their creative processes. There are ups and downs in building a company, but to see the work we're doing recognized as pioneering in this new age of software is a definitive highlight.

Just under two years ago, Invoke released one of the first Canvas interfaces for Stable Diffusion. Today, the team is launching the most significant update to Invoke since then: Invoke 5.0. This release introduces: - Control Canvas, a powerful new way to combine controlnets, IP adapters, regional guidance, inpainting, and traditional pixel-based drawing tools with layers on an infinite canvas. - Support for Flux models, including text-to-image, image-to-image, inpainting, and LoRA support. We’ll be expanding this in the coming weeks to include controls, IP adapters, and improved inpainting/outpainting. We’re also partnering with Black Forest Labs to provide commercial-use licenses to Flux in our Professional Edition, which you can learn more about at www.invoke.com. - Prompt Templates, making it easy to save, share, and re-use your favorite prompts Once again, we’re proud to be sharing these updates as OSS. You can download the latest release here: https://lnkd.in/egAE5Ytp or sign-up for the cloud-hosted version at www.invoke.com If you make anything cool/interesting, would love to see it. 🤘

If you're in VFX, I bet you know someone who will want to try this workflow out immediately. We hear a lot from VFX folks who are annoyed that they have to spend too much time going through large libraries of stock photography looking for a starting point image with the right feel and right depth to fit into the shot they are editing. Here’s a workflow for creating a depth-appropriate background shot using 2 reference images (1 for depth, 1 for style): 1 - Start with a reference image for depth. Pass it through ControlNet Depth Anything to capture the underlying structure. 2 - Take another reference image for style inspiration. Feed it into Global IP Adapter using style mode only. Adjust weight to your liking and pull down the end step %. 3 - Add text prompts to fine-tune the final output. This is only scratching the surface of the kind of control you can get in Invoke, especially once you start using models trained on your IP. #VFX #greenscreen #videoproduction #AI

You’ve made a great image generation workflow. But your team has no idea what IP Adapter and ControlNet mean. They shouldn’t have to, in order to get started. There are tons of great AI workflows out there using open source tech, but as a leader looking to scale, these tools won’t get adopted by enterprise teams if the whole team needs a glossary of AI terms and nodes to use them. I used Invoke’s workflow builder to create a simple app that only exposes the necessary fields (re-labeled for clarity). Here’s how it works: 1 - I experimented with settings in the studio to define all the settings and inputs that I wanted. 2 - I auto-generated a node-based workflow based on those settings 3 - I chose what input fields I wanted the simple application view to display, made a few tweaks, and re-labeled them to something more intuitive. 4 - *Done.* I have a workflow (essentially a mini-app) that I can share with others. Full control on the backend, simple/intuitive on the frontend, and an application that can grow with users as they get more familiar with the application. In this example, I used Ismail Seleit's FormFinder LoRA model to support this workflow, allowing an end-user to upload a sketch, a reference image, and a text prompt. The workflow runs those images through the appropriate models, with all the settings I’ve chosen, without the user having to adjust advanced settings like denoising strength or IP Adapter weights. This is how we make generative AI useful at scale - today, not tomorrow. #ai #aiarchitecture #aiproductivity #aiworkflows

As always, working with the Hugging Face team has been a pleasure - Don't think there is a bigger advocate or supporter of open source AI on the planet. Thanks to Julien Chaumond and team for the support in getting Invoke integrated.