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

Invoke
2 Apr 202406:43

TLDRInvoke 4.0 is released with a new model manager interface and support for evolving AI ecosystems. It allows easy model management, customization of default settings, and direct model installation from local paths or URLs. The conversion cache now stores up to 20GB of recent models. New features include trigger phrases for models and a gradient denoising feature for seamless generation. The release notes detail improvements, and there's an Nvidia AI workbench partnership for Invoke.

Takeaways

  • 🆕 **New Model Manager Interface**: Invoke 4.0 introduces a new model manager interface for easier management of models.
  • 💪 **Foundation for Future Evolution**: A strong foundation has been built to support the evolution of Invoke as the ecosystem evolves, including support for new model architectures.
  • 🔧 **Customizable Default Settings**: Users can now customize default settings for models and apply them system-wide with a single click.
  • 📁 **Enhanced Model Addition**: New models can be added via local paths, URLs, or directly from Hugging Face repositories, with options to select specific files.
  • 🔄 **In-Place Install Option**: Models can be installed in place, directly referencing files on the hard drive instead of copying them.
  • 🚀 **Conversion Cache**: Invoke now features a conversion cache to store recently used models in their converted state, speeding up repeated uses.
  • 🔄 **Non-Blocking Downloads**: Model downloads no longer block the user interface, allowing for multitasking.
  • 🗣️ **Trigger Phrases Feature**: Users can set trigger phrases for models to quickly access and use common prompts.
  • 🎨 **Gradient Denoising**: A new gradient denoising feature reduces generation time and eliminates seams in inpainting results.
  • 📝 **Unified Canvas Update**: The unified canvas has been updated to perform denoising in a single cycle, improving the generation process.

Q & A

  • What is the main feature introduced in Invoke 4.0?

    -Invoke 4.0 introduces a new model manager interface, which is part of a strong foundation for evolving Invoke as the ecosystem evolves.

  • What does the new model manager in Invoke 4.0 allow users to do?

    -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 models.

  • How does the new model manager help with customizing models in Invoke 4.0?

    -The new model manager allows users to set default settings for models, which can be easily updated and enabled. This is particularly helpful when switching between models that require specific parameters to be customized.

  • What is the benefit of the 'in place install' feature when adding new models?

    -The 'in place install' feature allows Invoke to directly reference the file on the user's hard drive instead of copying it into the Invoke model folder, saving space and making it easier to use the same model again.

  • How does the conversion cache work in Invoke 4.0?

    -Invoke 4.0 has a conversion cache that stores the most recent 20 GB of models in their converted state by default. This allows users to use the same model without having to convert it every time.

  • Can the conversion cache size be adjusted in Invoke 4.0?

    -Yes, the conversion cache size can be configured. Users can set it to zero if they do not want conversions to be stored on their machine, which would increase loading times but save hard drive space.

  • How has the installation process for new models improved in Invoke 4.0?

    -In Invoke 4.0, users can install multiple models at the same time without blocking the user interface. This allows for immediate work on models that are already on the machine while others are being downloaded.

  • 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 are compatible with the selected models. These can be easily accessed and used in the studio.

  • What is the new gradient denoising feature in Invoke 4.0?

    -The gradient denoising feature performs the denoising process in a blurred area around the mask, allowing for seamless inpainting results in a single generation cycle, significantly reducing the time to generate results.

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

    -Users can stay updated by checking the release notes, joining the Discord community, and looking at the invoke training repositories for information on the latest features and upcoming roadmap.

  • What is the partnership with Nvidia mentioned in the script?

    -Invoke has partnered with Nvidia to build the Nvidia AI Workbench project for Invoke. The 4.0 version of Invoke will be updated soon for this project.

Outlines

00:00

🚀 Introducing Invoke 4.0: Enhanced Model Manager and Features

Invoke 4.0 is announced with a new model manager and user interface. It's the beginning of a new era for Invoke, with a strong foundation built to support evolving ecosystems and new model architectures. The model manager allows for easy management of all models, including embeddings, VAEs, and control nets. It also features customization of default settings for models, which can be updated in the main studio. New models can be added via local paths, URLs, or directly from a Hugging Face repo. There's a new conversion cache that stores the most recent 20 GB of models to facilitate quick reuse. The video also mentions an 'in place install' feature, which references files directly on the hard drive instead of copying them. The new model manager also allows for installing multiple models simultaneously without blocking the user interface. Additionally, a 'trigger phrases' feature is introduced for main and concept models, which can be configured and reused in prompts.

05:00

🎨 Improved Performance and Seamless Integration in Invoke 4.0

The video highlights the significant reduction in time to generate inpainting results on the canvas with Invoke 4.0, thanks to a new gradient denoising feature. This feature allows for seamless image generation in a single cycle, leveraging technology by contributor dunon. The presenter also encourages viewers to check the release notes for a comprehensive list of updates in version 4.0. There's a mention of a partnership with Nvidia for the Nvidia AI Workbench project, with the 4.0 version to be updated soon. The video ends with a call to action to join Discord for more updates, to explore invoke training repositories for custom AI model creation, and to check out the new features for professional studios using Invoke's hosted product. The presenter also mentions the ability to customize prompt triggers for both stock and custom models.

Mindmap

Keywords

Invoke

Invoke is the name of the software discussed in the video, which appears to be an AI-based tool or platform. It is central to the video's narrative as the presenter discusses its new version, 4.0, and its various features. Invoke seems to be used for managing AI models and has undergone significant updates to enhance its capabilities.

Model Manager

The Model Manager is a new feature of Invoke 4.0 that allows users to manage different AI models from a single interface. It is highlighted as a significant improvement because it provides a more streamlined way to handle various models, including embeddings, VAEs, and control nets, directly from the panel.

User Interface

The term 'User Interface' refers to the point of interaction between the user and the system, in this case, the Invoke software. The script mentions a new model manager user interface, indicating that the look and feel, as well as the functionality, have been updated to improve the user experience.

Default Settings

Default settings in the context of Invoke 4.0 refer to the preset configurations for models that can be applied universally. The script explains that users can customize these settings and then apply them across all models, which is particularly useful for maintaining consistency when switching between different AI models.

Hugging Face Repo

A 'Hugging Face Repo' is mentioned as a source from which users can add new models to Invoke. It suggests that Invoke integrates with Hugging Face repositories, allowing users to directly access and utilize models hosted there, enhancing the software's versatility and model availability.

In-place Install

The 'In-place Install' feature mentioned in the script allows users to link directly to a model file on their hard drive rather than copying it into the Invoke model folder. This is beneficial for saving space and ensuring that any updates to the original file are reflected without needing to reinstall.

Conversion Cache

The 'Conversion Cache' is a new feature in Invoke 4.0 that stores the most recent models in their converted state, up to a certain size limit. This speeds up the process of using the same model multiple times by eliminating the need to convert it repeatedly.

Trigger Phrases

Trigger phrases are configurable prompts or terms associated with specific models in Invoke. They can be pre-set to quickly initiate certain model behaviors or to streamline the process of inputting common prompts, as demonstrated in the script where 'bold ink and watercolor' is set as a trigger phrase.

Unified Canvas

The 'Unified Canvas' seems to be a part of the Invoke interface where users can perform operations on models. The script discusses a new feature, gradient denoising, which is implemented on this canvas to improve the quality of model outputs by reducing visible seams in the generated content.

Gradient Denoising

Gradient Denoising is a technical feature that improves the visual output of the AI models by smoothing out the transitions in generated images. It is mentioned as a significant update in Invoke 4.0 that allows for a single-generation cycle to achieve seamless results, which was previously achieved through multiple generations.

Nvidia AI Workbench

Nvidia AI Workbench is mentioned as a partner platform with Invoke. The collaboration aims to make Invoke more accessible and integrated within Nvidia's AI development environment, suggesting that Invoke is expanding its compatibility and reach within the AI development community.

Highlights

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

A new model manager user interface has been introduced.

A strong foundation is built to support evolving model architectures.

The model manager allows direct management of all system models from a single panel.

Models can be customized with default settings for ease of use.

New models can be added via local path, URL, or directly from a Hugging Face repo.

The model manager includes a feature to scan and select individual files from a repo.

Local folders can be scanned to add default starter models.

In-place installation allows direct reference to files on the hard drive.

A conversion cache is introduced to store the most recent 20 GB of models.

The cache can be configured or set to zero for no additional hard drive space usage.

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

Trigger phrases can be configured for main models and concept models.

Trigger phrases can be quickly referenced and used in the studio.

A new gradient denoising feature has been added to the unified canvas.

The denoising process now occurs within a single generation cycle.

The new feature significantly reduces the time required for inpainting results.

Seams are minimized without the need for two generation passes.

Invoke has partnered with Nvidia for the Nvidia AI workbench project.

The 4.0 version will be updated soon for the Nvidia AI workbench.

Discord community is available for more information on the roadmap and upcoming features.

Training repositories are available for creating custom AI models.

Professional studios can customize prompt triggers for stock and custom models.

Stay tuned for more updates and features coming to Invoke.