Invoke - 4.0 Open Source Release - New Model Manager Interface, Gradient Denoising & More...

Invoke
2 Apr 202406:43

TLDRInvoke 4.0 has been released with a host of new features and improvements. The update introduces a new model manager interface, allowing users to manage and customize models directly from the panel. It supports new model architectures and offers a strong foundation for future evolution. The default settings can be updated for each model, and trigger phrases can be configured for quick access to common prompts. The conversion cache has been optimized to store only the most recent models, reducing hard drive space usage. The unified canvas now features a gradient denoising function that minimizes seams in a single generation cycle, significantly reducing inpaint generation time. The release notes provide a comprehensive list of updates, making Invoke 4.0 one of the most user-friendly versions yet. Users are encouraged to join Discord for more information on the roadmap and future developments.

Takeaways

  • 🚀 **New Model Manager Interface**: Invoke 4.0 introduces a revamped model manager with a user-friendly interface for managing models like embeddings, VAEs, and control nets.
  • 🔍 **Customizable Default Settings**: Users can now customize default settings for models, which can be applied across the main studio to streamline the customization process.
  • 📚 **Direct Model Installation**: New models can be installed directly from a local path or URL, including the ability to scan and select specific files from a repository.
  • 🔗 **In-Place Install Option**: A new feature that allows referencing files directly from the hard drive without copying them into the Invoke model folder.
  • 💾 **Conversion Cache**: Invoke now includes a conversion cache that stores the most recent 20 GB of models in their converted state to speed up usage.
  • ⚙️ **Configurable Cache Size**: Users can configure the cache size or set it to zero for no additional hard drive space usage, which increases loading time for unconverted models.
  • 🔁 **Multiple Model Installation**: Users can now install multiple models simultaneously without blocking the user interface, allowing for immediate work on already installed models.
  • 📈 **Trigger Phrases Feature**: A new feature that allows users to configure and store common prompts or phrases for easy reuse in the main models or concept models.
  • 🎨 **Gradient Denoising**: A new unified canvas feature that performs denoising in a blurred area around the mask, reducing the time for generating inpainting results.
  • 📝 **Release Notes**: For a full list of changes, the release notes should be consulted, which detail smaller updates contributing to the improved Invoke 4.0 experience.
  • 🤖 **Nvidia AI Workbench Partnership**: Invoke has partnered with Nvidia for the Nvidia AI Workbench project, with the 4.0 version soon to be updated for this project.
  • 🔄 **Custom AI Models**: Professional studios using Invoke's hosted product can now customize prompt triggers for stock models and any custom models uploaded into their account.

Q & A

  • What is the main focus of Invoke 4.0?

    -Invoke 4.0 introduces a new model manager and user interface, which is built on a strong foundation to support evolving model architectures and the ecosystem.

  • How does the new model manager in Invoke 4.0 help users manage their models?

    -The new model manager allows users to manage all models, including embeddings, VAEs, and control nets, directly from the panel. It also enables customization of default settings for main models.

  • What is the benefit of updating default settings in the new model manager?

    -Updating default settings allows users to set ideal parameters for specific models, making it easier to switch between models without any issues.

  • How can users add new models in the new model manager?

    -Users can add new models by pasting a local path or a URL, going directly to a Hugging Face repository, scanning a repository, or using the starter models tab.

  • What is the 'in place install' feature when adding models using a local path?

    -The 'in place install' feature allows the system to make a direct reference to the file on the user's hard drive instead of copying the files into the Invoke model folder.

  • How does the conversion cache in Invoke 4.0 work?

    -The conversion cache stores the most recent 20 GB of models in their converted state on the user's hard drive, making it easier to reuse the same model without having to convert it each time.

  • What is the trigger phrases feature in the model manager?

    -The trigger phrases feature allows users to configure and store common prompts or phrases that work with their main models or concept models for easy reuse.

  • What is the new gradient denoising feature on the unified canvas?

    -The gradient denoising feature performs the denoising process in a blurred area around the mask, which minimizes seams in a single generation cycle, reducing the time to generate an inpainting result.

  • How can users stay updated with the latest features and roadmap of Invoke?

    -Users can refer to the release notes for a full list of changes in 4.0, join the Discord for more information on the roadmap, and check out the invoke training repositories for creating custom AI models.

  • What is the partnership with Nvidia for Invoke?

    -Invoke has partnered with Nvidia to build the Nvidia AI Workbench project, which will soon be updated to include the 4.0 version of Invoke.

  • How can professional studios using Invoke's hosted product customize the prompt triggers for their models?

    -Professional studios can configure the prompt triggers for the stock models that come with Invoke Studio as well as for any custom models they upload into their account.

  • What is the benefit of configuring the model for the rest of the team in Invoke?

    -When a model is configured for the team, all team members can easily use the settings and prompt triggers with a click of a button, enhancing collaboration and efficiency.

Outlines

00:00

🚀 Introduction to Invoke 4.0's New Features

The video introduces the latest version of Invoke, version 4.0, which is a significant milestone in the platform's evolution. The update includes a new model manager and user interface, with a strong underlying foundation that supports evolving model architectures. The model manager allows for direct management of all models, including embeddings, VAEs, and control nets. It also features customizable default settings for main models, which can be updated and applied across the system. The video also discusses the new addition of a conversion cache, which stores the most recent 20 GB of models in their converted state to facilitate reuse without reconversion. The user interface has been improved to allow for non-blocking downloads of multiple models simultaneously. Additionally, a new trigger phrases feature has been added to the model manager, enabling users to configure and reuse common prompts for their models. Lastly, the video mentions a new gradient denoising feature on the unified canvas, which reduces the time required for inpainting results and eliminates noticeable seams.

05:00

📚 Release Notes and Future Collaborations

The speaker recommends viewers to check out the release notes for Invoke 4.0 to see a full list of changes and improvements. The release notes will cover smaller updates that collectively enhance the user experience. The video also announces a partnership with Nvidia to build the Nvidia AI Workbench project for Invoke, with an update for the 4.0 version expected soon. The speaker teases upcoming features and encourages the audience to join the Discord community for more information on the roadmap. For those interested in training, the speaker directs them to Invoke's training repositories to start creating custom AI models. The video concludes by highlighting that the changes discussed are already live, and users can now customize prompt triggers for stock and custom models in Invoke Studio, benefiting the entire team with a single configuration.

Mindmap

Keywords

💡Invoke 4.0

Invoke 4.0 refers to the latest version of the Invoke software, which is an open-source AI tool. It represents a significant update and is described as the beginning of a new era for the Invoke ecosystem. The update includes a new model manager interface and other enhancements that allow for better customization and management of AI models.

💡Model Manager Interface

The Model Manager Interface is a new feature in Invoke 4.0 that allows users to manage all their models, including embeddings, VAEs, and control nets, directly from a single panel. It is a user-friendly way to organize and customize the main models with default settings, making it easier to switch between different models.

💡Default Settings

Default settings in the context of Invoke 4.0 are pre-configured parameters for models that users can enable to optimize the model's performance for specific tasks. When updated, these settings are applied across the system to ensure the model operates at its ideal configuration.

💡Hugging Face Repo

A Hugging Face Repo is a repository where users can store and access machine learning models. Invoke 4.0 allows users to directly access models from a Hugging Face repository, making it easier to add new models to the system.

💡In-Place Install

In-Place Install is a feature that enables the Invoke software to reference a model file directly from its original location on the user's hard drive, rather than copying it into the Invoke model folder. This saves space and allows for direct updates to the model without the need for reconversion.

💡Conversion Cache

The Conversion Cache is a storage mechanism in Invoke 4.0 that holds the most recent converted models to speed up the process of using the same model repeatedly. It can be configured to store a certain amount of data, and users can adjust the cache size according to their storage preferences.

💡Trigger Phrases

Trigger Phrases are configurable prompts for specific models in Invoke 4.0. They can be pre-trained into the model or common prompts that users want to reuse. This feature allows for quick access and use of effective prompts when generating content with the selected model.

💡Unified Canvas

The Unified Canvas is a feature in Invoke 4.0 that provides a seamless interface for generating and editing AI models. It includes a new gradient denoising feature that improves the quality of image generation by reducing visible seams in the output.

💡Gradient Denoising

Gradient Denoising is a novel technology implemented in Invoke 4.0 that performs the denoising process in a blurred area around the mask. This feature allows for a single generation cycle to produce seamless inpainting results, significantly reducing the time required for image generation.

💡Nvidia AI Workbench

The Nvidia AI Workbench is a project partnership with Invoke, aimed at integrating Invoke into Nvidia's AI development environment. This partnership is expected to bring new features and improvements to the Invoke platform, especially for users working with Nvidia's AI tools.

💡Discord

Discord is a communication platform mentioned in the script where users can join to hear more about the roadmap and future updates for Invoke. It serves as a community hub for users to discuss, share ideas, and stay informed about the latest developments in the Invoke ecosystem.

💡Invoke Training Repos

Invoke Training Repos refer to the repositories where users can find resources and start creating their own custom AI models. These repositories are part of the Invoke platform and are designed to help users train and develop professional-grade AI models for various applications.

Highlights

Invoke 4.0 has been released, marking the beginning of a new era for the platform.

A new model manager user interface has been introduced in Invoke 4.0.

The update includes a strong foundation for evolving Invoke as the ecosystem evolves.

Support for new model architectures is built into Invoke 4.0.

Users can now manage all models, including lauras embeddings, VAEs, and control nets, directly from the model manager panel.

Main models can be customized with default settings, which can be updated and enabled for an ideal model setup.

Added models can be added via local path or URL, and even directly from a Hugging Face repository.

Invoke now supports installing models in place, making a direct reference to the file on the hard drive instead of copying.

A conversion cache is introduced to store the most recent 20 GB of models in their converted state for easy reuse.

The cache size can be configured, and the user can set it to zero for no additional hard drive space usage.

Multiple models can now be installed simultaneously without blocking the user interface.

Trigger phrases feature has been added for main models and concept models, allowing users to configure and reuse common prompts.

Unified Canvas now includes a gradient denoising feature that performs the denoising process in a blurred area around the mask.

The new feature reduces the time to generate an inpainting result on the canvas dramatically.

Invoke 4.0 release notes provide a full list of changes and updates.

A partnership with Nvidia has been announced for the Nvidia AI Workbench project.

Invoke training repositories are available for creating custom AI models.

Professional studios using Invoke's hosted product can now customize prompt triggers for stock and custom models.

Users can configure their model for the team, allowing others to use all settings and prompt triggers with a click.