AI Art Just Changed Forever

Theoretically Media
16 Nov 202313:03

TLDRThe video discusses a groundbreaking change in AI art generation with the introduction of Latent Consistency Models (LCMs) by Kaa, which allows for near-real-time image creation. The presenter explores Kaa's features, including the ability to use a painting or drawing program as input, manipulate generated images in real-time, and apply different styles. The video also covers Ever Art, an AI image generator that enables users to train their own models using uploaded images. The presenter shares their experience with Ever Art, demonstrating how it captures the style of the training images and can be influenced by additional prompts and reference images. The advancements in AI art generation are highlighted as exciting and hold great potential for future creative endeavors.


  • 🎨 A significant advancement in AI art creation has been introduced through Latent Consistency Models (LCMs), which allow for near real-time image generation.
  • πŸ–ŒοΈ LCMs can be integrated with painting or drawing programs, enhancing the creative process with features like shape and brush tools that react in real time.
  • πŸ“ˆ The technology is currently in beta, but the potential for widespread use is being explored, with plans to scale up access within a week.
  • 🌟 Users can experiment with styles, such as Cinematic and Illustrative, and manipulate generated characters in real time for dynamic posing.
  • πŸ”„ A 'randomized prompt' feature offers creative inspiration by generating different prompts and ideas for artists to explore.
  • πŸ–₯️ Image references can be used to guide the AI, allowing for unique interpretations rather than direct replicas of the source material.
  • πŸ”„ The ability to modify prompts and use an undo function provides flexibility and control over the final artwork.
  • πŸ“ EverArt, an image generator that allows users to train their own models, has a clean user interface and offers a straightforward training process for personalized styles.
  • πŸ“š Training a model with EverArt involves uploading up to 50 images, naming the model, and waiting for about 15 minutes for a fully trained model.
  • 🧩 Users can input their own information, like comic pages, to influence the style of generated images, creating unique and personalized outputs.
  • πŸ”— External software like Photoshop can be linked to the AI tool for a seamless workflow, although users should ensure they are in full-screen mode to avoid interface conflicts.
  • βš™οΈ The AI tool's capabilities are not limited to static images; it can also be used for digital sculpting and even real-time animation, showcasing the versatility of the technology.

Q & A

  • What is the major change in AI image creation mentioned in the transcript?

    -The major change is the introduction of latent consistency models (LCMs) which allow for near real-time image generation and can be used as an input in painting or drawing programs.

  • How does the LCM model enhance the user experience in image creation?

    -The LCM model enhances the user experience by generating images quickly and allowing users to manipulate and add details in real-time using various tools and features within a painting or drawing program.

  • What is the current status of the LCM feature?

    -The LCM feature is currently in beta, and the company is scaling up their GPU to handle more users without overloading the system.

  • What is Ever Art and how does it differ from the LCM model?

    -Ever Art is an AI image generator that allows users to train their own models by uploading a set of images. It differs from the LCM model in that it focuses on training models based on specific styles or themes to generate images.

  • How long does it take to train a model in Ever Art?

    -It takes about 15 minutes to train a model in Ever Art after uploading up to 50 images and submitting them.

  • What is one of the unique features of the AI image generator discussed in the transcript?

    -One unique feature is the ability to use image references or transparent PNGs to influence the generated image, as well as the option to link to an external screen for use with other painting software like Photoshop.

  • How does the AI react when the user makes changes to the generated image?

    -The AI reacts in real-time to changes made by the user, such as moving elements or altering the canvas, and it updates the image accordingly.

  • What is the significance of being able to link an external screen to the AI image generator?

    -Linking an external screen allows users who are more comfortable with specific painting software to integrate the AI's capabilities into their preferred workflow, enhancing flexibility and user experience.

  • What are some of the creative applications of the AI image generator mentioned in the transcript?

    -Some creative applications include digital sculpting in PlayStation software Dreams, real-time rendering in Blender with Pixar Animation style, and generating comic book illustrations influenced by specific comic styles.

  • How does the AI handle the addition of transparent PNGs to the generated image?

    -The AI can incorporate transparent PNGs into the generated image, allowing for the creation of more complex and layered visuals, such as adding characters or objects to an existing scene.

  • What is the process for using an LCM through Hugging Face?

    -The process for using an LCM through Hugging Face is outlined in another video by the same speaker, which is linked for further information.

  • What is the speaker's opinion on the AI's ability to create a one-to-one copy of a style?

    -The speaker prefers an AI-generated image that is influenced by a style rather than a one-to-one copy, as it allows for more creative and unique outputs.



🎨 Real-Time AI Image Generation with Latent Consistency Models

The speaker introduces a significant advancement in AI image creation with the advent of Latent Consistency Models (lcms). These models allow for near-instantaneous image generation, which can be further manipulated in real-time using a painting or drawing program. The feature is in beta, but the speaker has been informed that it will become widely available soon. The process involves setting a prompt, choosing a canvas fill color, and then using shapes and brush tools to modify the generated image. The AI responds quickly to these changes, allowing for dynamic adjustments. The speaker also discusses the ability to apply different styles, use randomized prompts for creative exploration, and pose characters by manipulating their shapes. Additionally, image references can be used to guide the AI's output, and there is an undo function for mistakes. The speaker concludes by noting that better artistry reduces the AI's workload.


πŸš€ Enhancing AI Art with Output Tweaks and External Software Integration

The speaker shares various tricks to enhance the AI-generated art. One method involves dragging an output over the base drawing to improve the result. Another fun feature is the addition of transparent PNGs, like Godzilla or explosion images, to create unique compositions. The speaker also mentions a new feature that allows linking to an external screen, enabling the use of familiar software like Photoshop for more comfortable and flexible editing. The video also covers the use of Kaa for digital sculpting and real-time rendering with Blender, showcasing the versatility of AI art tools. The speaker provides a timeline for when the public can expect access to Kaa's real-time generation capabilities and encourages signing up for the service. Lastly, the speaker revisits Ever Art, an image generator that allows users to train their models with their own images, and provides a detailed walkthrough of the process and results.


πŸ“š Training Ever Art with Personal Comic Book Style

The speaker discusses training Ever Art with personal comic book pages from 'Henchmen Inc', a comic that he had published and now has the rights to. He explains that uploading whole pages to Ever Art worked surprisingly well, and the AI was able to generate images that were stylistically accurate to the comic. The speaker also talks about using reference images to enhance the training process and provides examples of how Ever Art can be used with different prompts to generate images that are influenced by the inputted style. He concludes by expressing excitement about the increased control and flexibility in image generation and looks forward to seeing what the audience creates with these new tools.



πŸ’‘AI Art

AI Art refers to the creation of visual art through artificial intelligence. In the video, the host discusses how AI technology has recently advanced, allowing for real-time generation of images and art, which is a significant change in the field of digital art creation.

πŸ’‘Real-time Generation

Real-time generation is the process of creating or rendering images or animations on-the-fly as a user interacts with a system. In the context of the video, this refers to the ability of the AI to generate images as the user is painting or drawing, which is a new and exciting feature in AI art creation.

πŸ’‘Latent Consistency Models (LCMs)

Latent Consistency Models (LCMs) are a type of AI model that can generate images quickly, near real-time. They are highlighted in the video as a breakthrough technology that allows for the integration of AI with painting or drawing programs, enhancing the creative process.

πŸ’‘Canvas Screen

The canvas screen is the digital workspace where users can input prompts and start generating images. It is a key component of the AI art generation process shown in the video, where the user can set a prompt and begin to see AI-generated images as they work.


A prompt is a word or phrase that guides the AI in generating a specific type of image or art. In the video, the host uses prompts such as 'concept art sci-fi Planet' to direct the AI to create images that match those themes.

πŸ’‘Brush Tools

Brush tools are digital tools that simulate the effect of traditional painting brushes. They are used in the video to manually add details to the AI-generated images, allowing the user to interact with and modify the AI's output in real-time.


Styles in the context of the video refer to different visual themes or aesthetics that can be applied to the generated images. The host mentions 'Cinematic' and 'Illustrative' styles, which can alter the look of the AI-generated art to match specific visual styles.

πŸ’‘Randomized Prompt

A randomized prompt is a feature that automatically generates different prompts for the AI to create varied images. This is showcased in the video as a fun way to explore new ideas and concepts without manually setting prompts each time.

πŸ’‘Image References

Image references are existing images that are used as a source to guide the AI in generating new images. The host demonstrates this by using an image of Danielle van Deno as a pirate to influence the AI's output, showing how AI can adapt to given references.

πŸ’‘Ever Art

Ever Art is an AI image generator that allows users to train their own models using uploaded images. It is mentioned in the video as a tool that gives users a high degree of control and flexibility in generating images that are influenced by the style of the trained models.

πŸ’‘Digital Sculpting

Digital sculpting is the process of creating three-dimensional models using digital tools. In the video, it is mentioned that an artist has used AI technology for digital sculpting in the PlayStation software Dreams, showcasing the versatility of AI in art creation.


A major change has occurred in the way AI images and art can be created in real time.

LCMS (Latent Consistency Models) is a model that generates images quickly, near real time.

The model can be used with a painting or drawing program as an input for enhanced image generation.

The feature is currently in beta and expected to be widely available soon.

Users can set a prompt and start generating images with canvas fill color.

Shapes and brush tools can be used for dynamic and quick image creation.

The system allows for brush size and opacity control for detailed editing.

Different styles like Cinematic can be applied to the generated images.

A randomized prompt feature for exploratory phase to generate different ideas.

Real-time reaction to user's drawing adjustments, allowing for character posing.

Image references can be used to guide the AI in generating specific styles or characters.

The ability to modify the prompt to influence the generated image.

Outputs can be improved by dragging them over for secondary generation.

Transparent PNGs can be added to the generated images for additional effects.

External screen linking allows using preferred software like Photoshop with Kaa.

Kaa's real-time generation is expected to be accessible to more users within a week.

Ever Art is an image generator that allows users to train their own models with uploaded images.

Training a model in Ever Art is straightforward, taking about 15 minutes.

The flexibility in image generation has increased exponentially, offering more control to users.

Ever Art can generate images influenced by specific styles without needing a one-to-one copy.

The system can handle complex prompts and generate images with contextual similarity.