AI image generation glossary
Every term, explained in plain English
The vocabulary of AI image generation - negative prompts, seeds, CFG scale, img2img, samplers and more - defined clearly and accurately, with examples you can actually use.
Core concepts
6 terms- DenoisingDenoising is the core operation of a diffusion model: at each step it predicts and removes a little noise, gradually turning a random field into a clear image.
- GANA GAN (Generative Adversarial Network) generates images by pitting two neural networks against each other - one creating images, one judging them - until the fakes look real. It powered earlier AI image tools before diffusion models took over.
- Image-to-imageImage-to-image (img2img) is the AI workflow that transforms an existing picture according to your prompt, keeping some of the original structure instead of generating from scratch.
- img2imgimg2img is shorthand for "image-to-image" - the AI workflow that transforms an existing picture using a prompt, keeping some of the original structure instead of generating from scratch.
- Latent spaceLatent space is the compressed, abstract representation a diffusion model works in. Instead of manipulating millions of pixels, the model generates in this smaller space and then decodes it into an image.
- Text-to-imageText-to-image is the AI workflow where you type a written prompt and the model generates a brand-new image from it - no source image required.
Prompting
4 terms- Negative promptA negative prompt is a list of words describing what you do NOT want in the image. The model steers away from those concepts while it generates, which removes common flaws and unwanted elements.
- PromptA prompt is the written description you give an AI image generator telling it what to create. The clearer and more specific the prompt - subject, style, lighting, composition - the closer the result matches what you pictured.
- Prompt engineeringPrompt engineering is the practice of writing, structuring and refining prompts to reliably get the image you want - using description, emphasis, negative prompts and iteration instead of leaving the result to chance.
- Prompt weightingPrompt weighting lets you boost or reduce the influence of specific words in your prompt, so the model pays more attention to the parts that matter most to you.
Generation controls
6 terms- Aspect ratioAspect ratio is the proportional relationship between an image's width and height (for example 16:9 or 1:1). It defines the shape of the canvas and strongly influences composition.
- CFG scaleCFG scale (classifier-free guidance scale) controls how strongly the image follows your prompt. Low values are loose and creative; high values stick closely to the prompt but can look over-processed.
- ResolutionResolution is the pixel size of an image - its width by height. AI models generate best at a native resolution they were trained on; to go bigger and stay sharp you generate at a sensible size and then upscale.
- SamplerA sampler is the algorithm that decides how noise is removed at each step of generation. Different samplers reach the final image by different paths, trading off speed, detail and consistency.
- SeedA seed is the number used to initialize the random noise an image is built from. The same seed plus the same prompt and settings produces the same image every time, which makes results reproducible.
- StepsSteps (sampling steps) are the number of denoising passes the model runs to turn noise into an image. More steps can mean more detail, but past a point they only add time, not quality.
Editing & enhancement
5 terms- Hires fixHires fix (high-resolution fix) is a two-stage technique: generate at a safe native resolution, then upscale and run a second, partial diffusion pass so the larger image gains real detail without the artifacts of generating big in one go.
- InpaintingInpainting regenerates only a selected (masked) region of an image while leaving the rest untouched - useful for removing objects, fixing details or replacing part of a scene.
- OutpaintingOutpainting extends an image beyond its original edges, generating new content that continues the scene - used to widen a crop, change the aspect ratio, or build out a larger composition.
- Style transferStyle transfer applies the visual style of one source - an image, artist or aesthetic - onto your own content, keeping your subject while changing its look.
- UpscalingUpscaling increases an image's resolution. AI upscalers do more than stretch pixels - they intelligently add believable detail so the larger image stays sharp.
Models & training
8 terms- CheckpointA checkpoint is a saved AI model file containing the full set of trained weights. It is the complete "brain" that generates images - swapping checkpoints changes the entire look and capability.
- CLIPCLIP is a model trained to connect text and images in a shared space. In an image generator it acts as the text encoder, turning your prompt into the numbers that steer generation toward what your words mean.
- ControlNetControlNet is an add-on that conditions a diffusion model on a reference structure - such as a pose skeleton, edge map or depth map - so you control composition precisely, not just with words.
- Diffusion modelA diffusion model is the type of AI that powers most modern image generators. It learns to turn random noise into a coherent image by reversing a step-by-step noising process.
- DreamBoothDreamBooth is a fine-tuning method that teaches a base image model a specific subject - a person, pet, product or style - from just a few reference photos, so it can recreate that exact subject in new scenes.
- LoRAA LoRA (Low-Rank Adaptation) is a small add-on file that teaches a base model a specific style, character, object or concept - without retraining the entire model.
- Textual inversionTextual inversion teaches an AI image model a new concept - a style, object or character - and packs it into a single reusable keyword called an embedding, without altering the model's own weights.
- VAEA VAE (Variational Autoencoder) is the component that converts images between full-resolution pixels and the compressed latent space a diffusion model works in - encoding on the way in and decoding on the way out.
Licensing
1 termStop reading, start generating
The fastest way to understand any of these terms is to try them. Every PixelForge image ships with a commercial license - free tier included.
Open the generatorReady when you are
Start creating freeReady to get started?
Join thousands of creators using AI to generate stunning images.