The Complete Glossary of AI Image Terms (Prompts, Seeds, LoRAs, Upscaling & More)

This glossary breaks down the essential AI image terms in plain language, grouped so you can find what you need and actually use it.

Key takeaways:

• A prompt is your core instruction, and small wording changes can shift the whole image.
• Seeds control the random starting point, making outputs repeatable when you reuse them.
• LoRAs are lightweight add-ons that teach a model a specific style, character, or subject.
• Upscaling increases resolution and detail after generation.
• Samplers, steps, and CFG scale govern how the image is built and how closely it follows the prompt.

Prompt Fundamentals

Prompt. The text instruction you give the model describing what you want to see. It can range from a few words to a detailed paragraph covering subject, style, lighting, and composition. The prompt is the single most important input, and learning to write clear, specific prompts is the core skill of AI image generation.

Negative prompt. A separate instruction telling the model what to avoid, such as extra fingers, blur, or text. Negative prompts are a precise way to remove recurring problems without rewriting your main prompt.

Prompt weighting. Syntax that tells the model to emphasize or de-emphasize certain words, so a key element gets more attention than a minor one. This lets you fine-tune which parts of a prompt dominate the result.

Control and Reproducibility

Seed. The number that sets the random starting point for an image. The same prompt with the same seed and settings produces the same image, which makes seeds essential for reproducibility. Change the seed and you get a fresh variation; keep it fixed and you can tweak one detail while holding everything else steady.

Sampler. The algorithm that turns random noise into a finished image step by step. Different samplers produce slightly different looks and run at different speeds, so the choice affects both style and generation time.

Steps. The number of iterations the sampler takes to refine the image. More steps can mean more detail up to a point, after which returns diminish and you are just spending time.

CFG scale (Classifier-Free Guidance). A setting that controls how strictly the model follows your prompt. Low values give the model creative freedom; high values force it to adhere closely, sometimes at the cost of natural-looking results. Finding the right balance is key to getting what you asked for without artifacts.

Models and Customization

Base model (checkpoint). The large pre-trained model that does the actual image generation. Different base models have different strengths, aesthetics, and specialties, and switching between them changes the entire character of your output.

LoRA (Low-Rank Adaptation). A small, lightweight file that adjusts a base model to reproduce a specific style, character, object, or aesthetic without retraining the whole model. LoRAs are popular because they are quick to apply and easy to share, letting you layer a consistent look onto a general-purpose model.

Fine-tuning. The broader process of further training a model on a focused set of images so it specializes in a particular subject or style. Fine-tuning is heavier than using a LoRA but can produce deeper, more reliable results for a specific use case.

Embedding (textual inversion). A small trained file that teaches the model a new concept tied to a keyword, so you can summon a specific style or subject just by including that word in your prompt.

Quality and Output

Resolution. The pixel dimensions of the generated image. Higher resolution means more detail but also more compute, and many models have a native resolution they perform best at.

Upscaling. Increasing an image's resolution after generation, often using a dedicated AI upscaler that adds plausible detail rather than just enlarging pixels. Upscaling is how you take a fast, low-resolution draft and turn it into a high-resolution final asset.

Inpainting. Regenerating a selected region of an existing image while leaving the rest untouched, used to fix flaws or change specific elements. It is the precision-edit tool of AI image generation.

Outpainting. Extending an image beyond its original borders, generating new content that continues the scene. Useful for changing aspect ratios or adding context around a subject.

Img2img (image-to-image). Using an existing image as a starting point alongside your prompt, so the output is guided by both the reference and the text. This is the basis of reference-driven generation, where you steer new images toward an existing look.

Why Keeping Track of These Matters

The terms above are levers, and the settings that produce a great image, the exact prompt, seed, sampler, steps, CFG, and LoRA, are easy to lose the moment you move on to the next generation. Reproducing a winning result later is impossible if you cannot remember how you got it the first time.

This is where saving your prompts and settings as organized, reusable records pays off. A tool like MoonPrompt lets you store the full recipe behind a successful image, so you can rerun it, tweak one variable, or share it with your team without rebuilding the prompt from memory. As your library of working prompts grows, that organization becomes the difference between occasional lucky results and a dependable creative process.

Bringing It Together

You do not need to memorize every term to get great results, but knowing what each lever does turns guesswork into direction. Prompts set the vision, negative prompts remove problems, seeds and samplers control reproducibility and style, LoRAs and fine-tuning customize the model, and upscaling and inpainting refine the final output. Keep track of the settings that work, reuse them deliberately, and your AI image generation gets sharper and more consistent over time.

Frequently Asked Questions

What is a seed in AI image generation? A seed is the number that sets the random starting point for an image. Using the same seed with the same prompt and settings reproduces the same image, which is why seeds are essential when you want repeatable, controllable results.

What does a LoRA do? A LoRA is a small add-on file that adjusts a base model to reproduce a specific style, character, or subject without retraining the whole model. It is a lightweight, shareable way to layer a consistent look onto a general model.

What is the difference between resolution and upscaling? Resolution is the pixel dimensions an image is generated at, while upscaling increases that resolution after generation, often using an AI upscaler that adds detail rather than simply enlarging pixels.

What does CFG scale control? CFG scale controls how strictly the model follows your prompt. Lower values give the model more creative freedom, while higher values force closer adherence to the prompt, sometimes at the cost of natural-looking results.

‍