AI Art Terminology Glossary: 100+ Terms Explained

Your comprehensive reference for AI art and image generation terminology.

---

A

Aspect Ratio The proportional relationship between image width and height. Common ratios: 1:1 (square), 16:9 (widescreen), 9:16 (portrait/mobile).

Artifact Unwanted visual defects in generated images, such as distortions, noise patterns, or glitches.

Attention Mechanism The AI system's method of focusing on relevant parts of the prompt when generating specific image areas.

---

B

Batch Processing Generating multiple images simultaneously from the same or different prompts.

Bleeding When elements from different parts of a prompt or image unintentionally merge or overlap.

Bokeh The aesthetic quality of out-of-focus areas in an image, often achieved with prompt terms like "shallow depth of field."

---

C

CFG Scale (Classifier-Free Guidance) A parameter controlling how closely the AI follows your prompt. Higher values = stricter adherence; lower values = more creative freedom. Typical range: 7-12.

Checkpoint A saved state of a trained AI model. Different checkpoints produce different styles and qualities.

CLIP (Contrastive Language-Image Pre-training) OpenAI's model that understands relationships between images and text, used in many image generators.

Composition The arrangement of visual elements within an image—framing, balance, focal points.

ControlNet An add-on system for Stable Diffusion that provides structural guidance (poses, edges, depth) during generation.

---

D

DALL-E OpenAI's text-to-image AI model series. DALL-E 3 is the current version.

Denoising The process of removing noise from an image during generation, gradually revealing the final result.

Depth Map A representation of distance in an image, used by ControlNet to maintain spatial relationships.

Diffusion Model The AI architecture behind Stable Diffusion, Midjourney, and DALL-E. Works by gradually removing noise from random pixels to form images.

---

E

Embedding A trained concept or style saved as a small file, activated by specific trigger words in prompts.

Euler/Euler a A type of sampling algorithm used in image generation. "Euler a" is ancestral (adds randomness each step).

Exponential Moving Average (EMA) A technique for smoothing model weights during training, resulting in more stable outputs.

---

F

Few-Shot Learning Training AI on limited examples to learn new concepts or styles.

Fine-Tuning Adjusting a pre-trained model with additional training on specific data.

Flux A newer open-source AI image generation model known for prompt adherence.

---

G

GAN (Generative Adversarial Network) An older AI architecture where two networks compete—one generates, one evaluates. Precursor to diffusion models.

Generation A single image produced by the AI from a prompt.

Grid A multi-image output showing several variations from one prompt (common in Midjourney).

Guidance Scale See CFG Scale.

---

H

Hallucination When AI generates elements not specified in the prompt, sometimes adding unwanted details.

Hires Fix A technique to improve quality by generating at lower resolution first, then upscaling with AI enhancement.

Hyperparameter Settings that control how a model operates (learning rate, batch size, etc.).

---

I

Image-to-Image (img2img) Using an existing image as a starting point for generation, rather than starting from noise.

Inpainting Editing specific regions of an existing image while keeping the rest intact.

Interpolation Smoothly transitioning between two images or concepts.

---

K

Karras Scheduler A noise scheduling algorithm that often produces smoother, higher-quality results.

---

L

Latent Space The compressed mathematical representation where AI models manipulate image concepts.

LoRA (Low-Rank Adaptation) A lightweight method to customize models without full retraining. Adds specific styles, characters, or concepts.

LLM (Large Language Model) AI systems like GPT that understand and generate text, sometimes integrated with image generators.

---

M

Mask A selection defining which parts of an image to edit or protect.

Midjourney A popular commercial AI image generator known for artistic, aesthetic outputs.

Model Merge Combining multiple trained models to blend their capabilities and styles.

Multi-Modal AI systems that work with multiple types of input (text, images, audio).

---

N

Negative Prompt Instructions telling the AI what NOT to include in the image. Essential for quality control.

Noise Random pixel patterns that diffusion models start with and gradually refine into images.

NSFW Content filters limiting generation of adult or unsafe material.

---

O

Outpainting Extending an image beyond its original boundaries, generating new content that matches.

Overfitting When a model learns training data too specifically, losing ability to generalize.

---

P

Parameters Settings that modify how generation occurs (aspect ratio, steps, CFG, etc.).

Photorealistic Striving to look like a real photograph.

PMPT/Prompt The text description you provide to guide image generation.

Prompt Engineering The skill of crafting effective prompts to achieve desired results.

Prompt Weighting Assigning different importance levels to parts of your prompt.

---

R

Render The process of generating the final image, or a 3D-style generated image.

Resolution Image dimensions measured in pixels (e.g., 1024x1024).

Rollback Returning to a previous version or generation step.

---

S

Sampler/Sampling Method Algorithm determining how the AI removes noise during generation. Options include Euler, DPM++, DDIM, etc.

Seed A number that initializes the random generation process. Same seed + same prompt = same result.

Stable Diffusion Open-source text-to-image AI model by Stability AI, highly customizable.

Steps The number of denoising iterations during generation. More steps = higher quality (to a point).

Style Transfer Applying the artistic style of one image to the content of another.

SDXL (Stable Diffusion XL) The latest major version of Stable Diffusion with improved quality and resolution.

---

T

Text Encoder The component that converts text prompts into numerical representations the AI can understand.

Text-to-Image (txt2img) Generating images from text descriptions alone.

Token Individual units of text that prompts are broken into for processing. Most prompts have a token limit.

Training Data The images and text pairs used to teach the AI model.

---

U

Upscaling Increasing image resolution, often with AI enhancement to add detail.

U-Net The neural network architecture at the core of many diffusion models.

---

V

VAE (Variational Autoencoder) The component that encodes/decodes images to/from latent space. Different VAEs affect color and sharpness.

Variation A new image generated from an existing one, similar but not identical.

ViT (Vision Transformer) A neural network architecture for processing images, used in CLIP and other models.

Volumetric Lighting Light rays visible in atmosphere (god rays, light beams through fog).

---

W

Weights The numerical values in neural networks that determine outputs. Also, emphasis values in prompts.

Wildcard A placeholder in prompts that randomly selects from predefined options.

---

X

X/Y/Z Plot A grid comparing different parameter values to see their effects.

XL Designates higher-capability model versions (e.g., SDXL, Leonardo XL).

---

Z

Zero-Shot Generating from a description without any task-specific training examples.

---

Common Acronyms Quick Reference

---

Conclusion

This glossary covers the essential terminology for AI image generation. Bookmark this page and refer back as you encounter new terms in your AI art journey!

Tip: Use our Image to Prompt tool to analyze images and learn which techniques and styles were used to create them.