pages.resourcesGlossary.hero.title

Adobe's family of generative AI models integrated into Creative Cloud applications. Notable for being trained exclusively on licensed and public domain content, addressing copyright concerns that surround other AI tools.

An emerging paradigm where AI systems operate with greater autonomy in creative processes, making decisions about composition and style with minimal human guidance. This raises fundamental questions about artistic intention and authorship.

Documentation and tracking of how AI-generated artwork was created, including model, prompts, and settings. Standards like C2PA aim to establish transparency about AI involvement, increasingly seen as essential for ethical practice.

A neural network component that allows the model to focus on the most relevant parts of its input when producing output. Cross-attention between text and image features is central to how text-to-image models interpret prompts.

The application of diffusion model principles to audio generation, treating sound as a signal to be denoised from randomness. It enables more coherent and higher-quality AI music and sound effects than earlier methods.

A saved snapshot of a trained AI model at a particular point in its training process. Different checkpoints produce dramatically different visual styles, and artists often switch between them depending on the desired output.

A parameter controlling how strongly a diffusion model adheres to the text prompt during image generation. Higher values produce closer prompt matches but may sacrifice naturalness; lower values give more creative freedom.

A neural network by OpenAI that associates images with text descriptions, serving as the bridge between language and vision in many AI art systems. Trained on hundreds of millions of image-text pairs.

A node-based graphical interface for building AI image generation workflows. Unlike simpler interfaces, ComfyUI exposes the full pipeline as a visual graph, giving users granular control over every step of the generation process.

A neural network architecture that adds spatial conditioning to diffusion models, allowing precise control over the structure of generated images using edge maps, depth maps, or pose skeletons. It bridges freeform generation and intentional design.

The legal right granting creators exclusive control over reproduction and distribution of their original works. In AI art, key questions include whether AI outputs can be copyrighted and whether training on copyrighted works infringes creators' rights.

A set of standardized licenses allowing creators to grant specific usage permissions while retaining some rights. The use of CC-licensed work in AI training datasets is a contested ethical issue in the creative community.

A series of text-to-image AI models by OpenAI. DALL-E 3 significantly improved prompt adherence and text rendering, and is integrated into ChatGPT and Microsoft products, making it one of the most widely accessible AI image generators.

Synthetic media created by AI that convincingly replaces a person's likeness, voice, or actions with fabricated content. Deepfakes raise serious ethical concerns around consent, misinformation, and identity, prompting legislative action worldwide.

A class of generative model that learns to create images by gradually removing noise from a random starting point. During training, the model learns how noise is progressively added to images; during generation, it reverses this process to produce coherent visuals.

A numerical vector representation of a concept, word, or image in a continuous mathematical space. Custom embeddings allow artists to teach AI models new concepts using just a few example images through textual inversion.

A legal doctrine in U.S. copyright law that permits limited use of copyrighted material without permission. Whether AI training on copyrighted images constitutes fair use is one of the most consequential unresolved legal questions in the AI art space.

The process of further training a pre-trained AI model on a specific dataset to adapt it for a particular task or style. It allows artists to teach a model their unique aesthetic using less data and computation than training from scratch.

A family of open-source image generation models by Black Forest Labs that introduced architectural innovations improving image quality, prompt adherence, and text rendering. It represents the cutting edge of open-source generative art.

A neural network architecture consisting of two competing networks: a generator that creates synthetic data and a discriminator that distinguishes real from fake. Through adversarial training, the generator produces increasingly realistic outputs.

Artificial intelligence systems that can create new content such as images, music, text, or video based on patterns learned from training data. Unlike discriminative AI that classifies inputs, generative AI produces novel outputs that did not exist before.

When an AI model generates content that is plausible-looking but factually incorrect or nonsensical. In images, this manifests as anatomical errors, impossible physics, or garbled text. While sometimes creatively serendipitous, it is generally a limitation.

A generation mode where an existing image serves as the starting point for AI transformation. The model uses the input image as a structural guide while applying changes based on a text prompt, ranging from subtle edits to radical reimaginations.

An AI-powered editing technique that fills in selected regions of an image with new content that seamlessly blends with the surrounding context. Artists use inpainting to remove unwanted elements, repair damage, or replace portions of an image.

A compressed, abstract mathematical representation of data learned by a neural network. In generative models, images are encoded as points in latent space where nearby points correspond to visually similar outputs, enabling creative exploration.

A fine-tuning technique that customizes large AI models by training only a small number of additional parameters. LoRAs are compact files that can be shared and applied to base models to add specific styles, characters, or concepts.

A commercial AI art generation platform known for producing highly aesthetic, stylized imagery. Accessible primarily through Discord, it has become one of the most popular text-to-image tools among artists and designers.

A phenomenon where AI models trained on AI-generated content progressively degrade in quality and diversity. As synthetic images populate the internet, future models risk training on them, creating a feedback loop that highlights the value of human-created art.

A text instruction that tells an AI model what to avoid or exclude from the generated output. While the positive prompt describes desired content, the negative prompt steers the model away from unwanted artifacts, styles, or elements.

A computational architecture inspired by biological neurons forming the foundation of modern AI. Deep neural networks with many layers enable the complex pattern recognition and generation capabilities used in creative AI applications.

An AI technique that reconstructs 3D scenes from 2D photographs by learning a continuous volumetric representation. NeRFs bridge photography and 3D art with applications in virtual production, game design, and immersive installations.

The predefined sequence determining how noise is added to and removed from images during diffusion. It controls the pace of generation, affecting how quickly the model commits to broad structures versus fine details.

A mechanism allowing artists to request exclusion of their work from AI training datasets. Various opt-out systems exist, from robots.txt to dedicated registries, but their effectiveness versus opt-in approaches remains debated.

The process of using AI to extend an image beyond its original boundaries, generating new content that continues the scene in any direction while maintaining consistency in style, perspective, and lighting.

Algorithmic creation of content using rules, parameters, and randomness rather than manual design. While predating modern AI, combining procedural methods with neural networks represents a powerful frontier in computational creativity.

The practice of crafting effective text instructions to guide AI models toward producing desired outputs. It involves understanding how models interpret language, using specific artistic vocabulary, and iteratively refining instructions.

Creative works not protected by intellectual property laws that can be freely used by anyone. Works enter the public domain when copyright expires or creators waive rights. Public domain artwork is increasingly used as ethically uncontested AI training data.

A creative AI company offering tools for video generation, image editing, and multimedia creation. Its Gen-2 and Gen-3 models pioneered accessible text-to-video generation for film, advertising, and content creation.

The process by which a generative model produces an output from its learned distribution. Different sampling algorithms offer tradeoffs between speed, quality, and determinism, giving artists control over generation characteristics.

An open-source text-to-image diffusion model whose open weights and permissive licensing spawned a vast ecosystem of community modifications and custom interfaces. It is the foundation of the open-source AI art movement.

An AI technique that isolates individual instruments or vocal tracks from mixed audio recordings using neural networks. It has transformed music production, DJing, and archival restoration by enabling manipulation of previously inseparable elements.

An AI technique that applies the visual style of one image to the content of another, such as rendering a photograph in the style of Van Gogh. Neural style transfer was one of the earliest mainstream applications of AI in visual art.

An AI music generation platform that creates full songs with vocals, instrumentation, and lyrics from text prompts. It represents a significant leap in AI music quality, sparking debate about the future of music composition.

A category of AI systems that generate visual images from natural language descriptions. The user provides a text prompt and the model produces a corresponding visual output, democratizing image creation while raising questions about authorship.

AI systems that generate musical compositions from natural language descriptions of genre, mood, instrumentation, and tempo. Platforms like Suno and Udio have made it possible to create full songs from text prompts.

The basic unit of text that AI models process, typically representing a word or subword. Understanding tokenization helps artists write more effective prompts by knowing how models parse their instructions.

The component that converts raw text into numerical tokens an AI model can process. Understanding tokenization explains why certain phrasings produce better results and why some concepts are split across multiple tokens.

The dataset used to teach an AI model patterns, styles, and concepts. For image models, training data typically consists of millions of image-text pairs, and its composition is at the center of ongoing ethical and legal debates.

A neural network architecture based on self-attention that has revolutionized language and image AI. Transformers process data in parallel, capturing long-range relationships, and form the backbone of large language models and modern image generators.

The process of increasing an image's resolution while adding plausible detail using neural networks. AI upscaling produces results far superior to traditional interpolation and is a standard step in AI art workflows.

A neural network that compresses images into compact latent representations and reconstructs them. In diffusion pipelines, the VAE encodes images for efficient processing and decodes the final result back to full resolution.

AI technology that replicates a person's voice from audio samples, enabling generation of new speech in that voice. In music, it has been used to create AI vocals mimicking famous artists, raising ethical and legal concerns.