5 Workflows to Create AI Images That Don't Look Generic

Invoke
18 Apr 202455:16

TLDRThis workshop focuses on innovative methods for creating unique AI-generated images, particularly through the lens of monster creation. The speaker illustrates various techniques for devising non-generic AI monsters, starting with basic sketches and evolving them through text-to-image AI tools. Techniques include manipulating image aspect ratios, integrating distinct traits like lizard scales, and refining these creations with specific attributes such as horns or unique textures. The session emphasizes the importance of creativity and making distinct artistic decisions to steer clear of generic AI art, showcasing a methodical approach to developing unique creature concepts.

Takeaways

  • 🎨 **Embrace Non-Generic Creation**: The speaker emphasizes the importance of creating unique and non-generic AI images, especially in the context of designing monsters and creatures.
  • 🚀 **Start with Crazy Ideas**: The process begins with brainstorming wild concepts, especially when designing something that doesn't exist, like mythical creatures.
  • 📝 **Use Prompts Effectively**: The script discusses using prompts to guide the AI in creating images, but also infusing different concepts to avoid generic results.
  • 🌟 **Iterate and Refine**: The creative process involves taking a basic idea, iterating over it, and refining it to achieve a unique and compelling result.
  • 🔄 **Control the AI with Initial Images**: The speaker demonstrates using an initial image to guide the AI towards a specific structure while allowing room for creative interpretation.
  • ✍️ **Incorporate Artistic Input**: Adding manual artistic input, like drawing additional elements on the canvas, allows for further customization and uniqueness.
  • 🧩 **Combine Techniques**: The script suggests combining different techniques, like using control nets and image-to-image transformations, to achieve complex and non-generic designs.
  • 🎭 **Inspire with Thematic Changes**: Changing the theme of a sheet of creatures by using an IP adapter can inspire new ideas and directions for creature design.
  • 🔍 **Focus on Details**: The importance of focusing on details and making creative decisions about the aspects that will make the creation stand out is highlighted.
  • 🧵 **Connect the Dots**: The process is likened to playing with puzzle pieces, where useful elements are combined and refined to form a coherent whole.
  • 📈 **Regional Control for Advanced Creation**: A sneak peek at an upcoming tool for regional control is given, promising even greater precision and creativity in AI image generation.

Q & A

  • What is the main theme of today's event?

    -The main theme of today's event is to explore ways to create AI images that are non-generic, focusing on the concept of monster creation.

  • What technique did the speaker use to generate initial monster ideas?

    -The speaker used a technique involving generating a sheet of monsters and selecting elements they liked from it to create unique non-generic creatures.

  • How can the aspect ratio of an image be manipulated during the AI image creation process?

    -The aspect ratio can be manipulated by using the 'lock that aspect ratio' feature and adjusting the image size to make it larger or smaller as desired.

  • What role does the 'prompt' play in the AI image creation process?

    -The 'prompt' guides the AI in infusing different concepts and character traits into the generic structure of the image, allowing for more creative and unique outputs.

  • How can adding elements to the canvas help in the creative process?

    -Adding elements to the canvas allows for reimagining certain parts of the image. It enables the artist to modify and detail specific aspects, such as adding horns or other features to the creature.

  • What does the speaker mean by 'non-normative' in the context of art creation?

    -In the context of art creation, 'non-normative' refers to the intention to deviate from what is considered normal or expected, aiming to create something different and unique.

  • How does the speaker describe the process of making creative decisions in art?

    -The speaker describes the process as one of making decisions at each point to be more or less like what's out there, pushing towards or away from certain aspects to create something non-generic.

  • What is the significance of using an initial image to shape the noise in AI art creation?

    -Using an initial image to shape the noise allows the artist to guide the AI towards a specific direction, creating a base that can be built upon and manipulated to achieve the desired outcome.

  • How can the AI model be influenced to generate more non-generic images?

    -The AI model can be influenced by using control adapters, regional prompting, and providing it with examples or references that deviate from the norm, thus pushing it towards more unique and creative outputs.

  • What is the challenge when trying to create non-anatomical human forms with AI?

    -The challenge is that AI models are often trained on data that assumes a certain norm, like humans having two arms. To create non-anatomical forms, the model needs to be guided or trained to understand and generate these non-standard anatomies.

  • What upcoming tool was hinted at that could provide more control over AI image creation?

    -The upcoming tool hinted at is 'Regional prompting control' or 'Regional control,' which allows for more precise control over different parts of the generated image.

Outlines

00:00

🎨 Exploring Non-Generic Monster Creation

The video discusses the importance of creating non-generic content, particularly in the context of designing monsters and creatures. It emphasizes the need to avoid clichés and to infuse unique concepts into the creative process. The speaker shares a technique for generating ideas by starting with basic shapes and then adding distinctive traits using prompts. The process involves using an image-to-image transformation tool to create a 'Tiger Raptor' creature, adjusting the prompts to control the outcome, and manually editing the final design for a more original result.

05:00

🛠️ Creative Decision Making in Art

This paragraph delves into the concept of creative decision-making in the arts. It highlights that every artistic choice is a decision to align with or deviate from the norm. The speaker discusses the challenge of being non-normative while using AI art, which tends to produce generic results if not guided properly. The video demonstrates how to use an image-to-image tool to create unique creatures by starting with a base image and iteratively refining it through creative decisions, resulting in a horned Raptor-tiger creature.

10:02

🧩 Iterative Refinement and AI Influence

The speaker talks about the iterative process of refining AI-generated images to achieve a desired outcome. They discuss the use of control images and prompts to guide the AI in generating specific features, such as adding horns to a creature. The paragraph also touches on the idea of using pairs of elements simultaneously for consistency and the potential need for manual touch-ups in the creative process. The video ends with the creation of a unique 'radillo' creature, showcasing the iterative approach to AI art.

15:04

🤔 Pushing AI Models Towards Non-Generic Creations

This section addresses the challenge of using AI art to create non-generic content when the AI model is trained on a vast amount of data, which can lead to average-looking results. The speaker suggests using initial images and customizing the AI model to push it towards more unique creations. They demonstrate this by using an image of a crustacean to influence the generation of a new creature, showing how different inputs can lead to diverse and interesting outcomes.

20:07

🎭 Adding Details and Embellishments to AI Creations

The paragraph focuses on adding details and embellishments to AI-generated creatures to make them more unique and interesting. The speaker describes how to use a control adapter to maintain the structure of the creature while adding features like feathers. They also discuss the idea of using snapshots and IP adapters to maintain consistency in certain features across generations. The video shows the evolution of a creature with vibrant feathers, highlighting the artist's influence in guiding the AI.

25:08

🧬 Experimenting with Control Adapters and Image Prompts

The speaker shares their process of experimenting with control adapters and image prompts to create unique creatures. They demonstrate how to use a control image to maintain the overall structure while introducing new visual elements through image prompts, such as the sheen of a hagfish or the texture of a crawfish. The video shows the generation of a creature with transparent spikes, which is then refined and detailed to create a non-generic monster.

30:19

🔍 Deep Sea Creatures and Non-Generic Inspiration

In this part, the speaker explores the creation of deep sea creature-inspired monsters, emphasizing the importance of non-generic ideas. They discuss the use of control images and the potential for unexpected results when combining different prompts. The video shows the transformation of a creature with extra legs and spikes, which are then refined to create a unique 'deep sea fake' creature.

35:21

🚀 Creating Non-Anatomical Humans and Monstrous Features

The paragraph explores the creation of non-anatomical human characters, such as a character with four arms. The speaker discusses the challenges of guiding the AI to understand and generate non-standard human anatomy. They demonstrate the process of manually editing the AI's output to achieve the desired result and mention the potential of future tools like regional prompting to better control non-generic features.

40:23

📈 Training AI Models for Non-Generic Outputs

The speaker talks about the importance of training AI models to recognize and generate non-generic content. They discuss the concept of 'aura' training, where multiple examples of a non-standard feature are provided to the AI to help it understand and replicate that feature. The video concludes with a sneak peek at a new tool for regional control, which promises to give artists more precise control over AI-generated images.

Mindmap

Keywords

💡Non-generic

The term 'non-generic' refers to something that is not typical or ordinary and stands out from the norm. In the context of the video, it is used to describe the process of creating unique and original AI-generated images that do not resemble common or clichéd representations. The video emphasizes the importance of avoiding generic ideas in favor of more innovative and distinctive concepts, such as designing monsters and creatures with unusual features.

💡Monster creation

Monster creation is the process of designing and conceptualizing fictional creatures that are often used in various forms of media like games, movies, and literature. The video discusses this concept as a way to explore non-generic ideas, where the creator 'throws crazy ideas at the wall' to generate unique and imaginative monsters that don't exist in reality.

💡Image to image

Image to image is a technique used in AI image generation where an initial image is transformed into another image based on a given prompt or set of instructions. In the video, this technique is used to evolve a basic concept into a more complex and detailed creature, guiding the AI to create something non-generic by starting with a simple structure and infusing it with various concepts.

💡Control net

A control net is a tool used in AI image generation that allows the user to guide the AI's output by providing a reference image. The video demonstrates using a control net to maintain the structural integrity of the original image while allowing for the infusion of new concepts, such as adding horns or other features to the creature being designed.

💡Creative decision making

Creative decision making involves the process of making choices that determine the direction and nature of a creative work. The video emphasizes that art, including AI-generated art, is fundamentally about making decisions. It discusses how creators can use AI tools to make more or less normative decisions, pushing the boundaries of what's expected to create non-generic outcomes.

💡Denoising strength

Denoising strength is a parameter in AI image generation that controls the level of detail and noise reduction in the final output. A higher denoising strength results in a cleaner, more refined image, while a lower strength allows for more variability and 'noise'. In the video, adjusting denoising strength is part of the process to achieve the desired look for the AI-generated creatures.

💡Regional prompting

Regional prompting is an advanced feature in AI image generation that allows for control over specific parts or 'regions' of the generated image. The video provides a sneak peek at this upcoming tool, which promises to give users an unprecedented level of control by allowing them to apply different prompts to different areas of the image, potentially leading to more complex and non-generic creations.

💡AI training

AI training refers to the process of teaching an AI model to recognize and generate specific features or concepts by providing it with numerous examples. In the context of the video, training an AI model like DALL-E with various examples of non-generic features, such as four-armed humans, can help the model better understand and generate such concepts when prompted.

💡Concept painting

Concept painting is an artistic process used in the early stages of creating visual media, such as films or video games, to visualize and develop ideas for characters, environments, and other elements. The video discusses using AI-generated images in a similar fashion, starting with a basic concept and iteratively refining it to create a unique and compelling final design.

💡Normative

Normative refers to something that is based on or conforms to a standard, typical pattern, or expectation. In the video, the term is used to contrast with 'non-generic', highlighting the challenge of guiding AI to create images that deviate from the norm. The video discusses how AI tends to generate outputs that align with its learned 'norms', making it harder for it to produce truly non-generic content without explicit guidance or examples.

💡Iteration

Iteration is the process of repeating a process with the aim of improving or refining the output. In the context of the video, iteration is a key part of the creative process when using AI to generate images. The creator takes the initial AI-generated image and makes successive refinements, each time adjusting and guiding the AI to move closer to the desired non-generic concept.

Highlights

The event focuses on creating non-generic AI images, particularly in the realm of monster creation.

The presenter discusses the importance of avoiding generic outputs when generating AI images.

A technique for generating ideas is demonstrated using a sheet of monsters as a starting point.

The use of image-to-image transformation to create a 'Tiger Raptor' creature is shown.

The presenter emphasizes the role of creative decision-making in the art creation process.

An example of adding horns to a creature and adjusting its features live is provided.

The concept of using an 'IP adapter' to influence the direction of AI-generated images is introduced.

A method for creating a 'Zebra Manticore' with vibrant feathers is explored.

The challenges of generating non-anatomical human features, such as four arms, are discussed.

The presenter illustrates how to manually adjust and 'paint out' additional arms in a generated image.

The limitations of AI models when faced with non-normative prompts are highlighted.

A sneak peek at an upcoming tool for regional control in AI image generation is given.

The potential of regional prompting for creating non-generic and highly controlled AI images is discussed.

The presenter expresses excitement about the new level of control and creative possibilities the new tool will offer.

The session concludes with a Q&A addressing the training of AI models to understand non-normative concepts.

The importance of showing the AI examples of non-standard features, like fingers for hair, is emphasized for training purposes.

The presenter assures that the new regional control tool will be integrated with existing controls for a seamless workflow.