CropUp Ai
When you first start using AI image generation tools the terminology can feel overwhelming.
Prompts. Parameters. Seed numbers. Stylize values. Omni references. Negative prompts. Upscaling. Inpainting. LoRA. CFG scale. The vocabulary accumulates fast and most resources assume you already know what these things mean.
This glossary defines every term you will encounter as a beginner - in plain English, with practical context for how each one affects your work. Bookmark it. You will come back to it.
Core concepts
Prompt
The text description you write to tell an AI image generator what to create. The prompt is your primary creative tool. The quality and specificity of your prompt determines the quality and accuracy of the output. A well-structured prompt follows the 6-layer framework: content type, subject, colors and materials, composition, lighting, and camera and lens.
Negative prompt
An instruction telling the AI what NOT to include in the image. In Midjourney negative prompts use the --no parameter. In ChatGPT they are written into the main prompt as exclusion language. Negative prompts solve specific recurring problems - unwanted text appearing in backgrounds, extra hands in product shots, generic AI aesthetics you want to avoid.
Parameter
A modifier added to the end of a Midjourney prompt using two dashes (--). Parameters control specific technical aspects of the generation - aspect ratio (--ar), model version (--v), stylization level (--stylize), and many others. Parameters give you precise control over output characteristics that you cannot control through descriptive language alone.
Generation
A single instance of running a prompt through an AI image generator. Each generation produces one or more output images. In Midjourney each generation produces a 2x2 grid of 4 image variations. Running a prompt is called generating or running a generation.
Iteration
The process of refining a prompt through multiple generations. You generate, evaluate the output, adjust the prompt based on what did and did not work, and generate again. Iteration is the core workflow of AI image production. Professional results almost always require multiple iterations.
Model
The underlying AI system that generates images. Different models produce different aesthetic qualities and have different capabilities. Midjourney V8.1, V8, and V7 are different models. ChatGPT uses DALL-E. Choosing the right model for your use case significantly affects output quality.
Training data
The images used to teach an AI model what things look like. A model trained on millions of photographs learns to generate photographic images. The training data determines what a model knows how to render and what it does not. This is why AI models sometimes struggle with very specific or niche subjects.
Midjourney specific terms
--ar (aspect ratio)
Sets the dimensions of the output image. --ar 1:1 produces a square. --ar 4:5 produces a portrait rectangle suited for Instagram. --ar 16:9 produces a wide landscape. Always set this before generating to avoid cropping issues later.
--v (version)
Specifies which Midjourney model version to use. --v 8.1 is the latest and most capable as of 2026. --v 7 is the previous generation. Different versions produce different aesthetic qualities.
--stylize
Controls how much creative interpretation Midjourney applies to your prompt. Range is 0 to 1000. Low values (0 to 100) follow your prompt very literally. High values (500 to 1000) apply strong artistic interpretation that drifts from your exact description. Default is 100.
--chaos
Controls how different the 4 images in each generation grid are from each other. Low chaos produces 4 similar variations. High chaos produces 4 dramatically different interpretations. Range is 0 to 100. Default is 0.
--raw
Disables Midjourney's built-in aesthetic filter. Without --raw Midjourney automatically applies its own sense of polish and beauty. With --raw it follows your prompt more literally without enhancement. Useful for photorealistic and documentary-style content.
--seed
A number that makes generations reproducible. The same prompt with the same seed produces the same image every time. Useful for replicating specific results or sharing reproducible prompts with others.
--no
The negative prompt parameter. Everything listed after --no is excluded from the generation. Example: --no text, watermarks, people, extra fingers.
--draft
Generates images at lower quality using fewer credits. Use draft mode to test prompt directions before committing to full quality generations. Approximately one fifth of the credit cost of a full generation.
--weird
Introduces unusual and unexpected elements into the generation. Range is 0 to 3000. Higher values produce increasingly surreal results that deviate from conventional aesthetics.
--sref (style reference)
Transfers the aesthetic style of a reference image to your generation. Midjourney analyzes the visual style - color palette, lighting quality, texture, mood - and applies those qualities to your output. Does not transfer specific subjects from the reference image. Used with an image URL: --sref [URL].
--oref (omni reference)
Maintains the visual identity of a specific subject across multiple generations. Available in V7 and later. Upload a reference image of your product or character and Midjourney maintains its visual identity across different scenes. Used with an image URL: --oref [URL].
--p (personalization)
Activates your personalization profile. Midjourney applies your trained aesthetic preferences to the generation. Requires completing a minimum number of image rankings to build the profile before --p is usable.
--sw (style weight)
Controls the strength of a style reference applied with --sref. Range is 0 to 1000. Higher values apply the reference style more strongly.
--ow (omni weight)
Controls the strength of an omni reference applied with --oref. Range is 0 to 1000. Higher values maintain the reference subject more closely.
U buttons (U1, U2, U3, U4)
Upscale buttons in the Midjourney Discord interface. After a generation produces a 2x2 grid of 4 images, clicking U1 through U4 upscales the corresponding image to full resolution. U1 is the top left image, U2 is top right, U3 is bottom left, U4 is bottom right.
V buttons (V1, V2, V3, V4)
Variation buttons in the Midjourney Discord interface. Clicking V1 through V4 generates 4 new variations based on the corresponding image from the previous grid. Use these to explore variations on a direction you like without changing the core concept.
Remix mode
A Midjourney setting that allows you to modify your prompt when generating variations. Without remix mode variations use your original prompt. With remix mode you can change specific elements of the prompt while keeping the overall composition and style.
Personalization profile
A trained model of your aesthetic preferences built through the image ranking process. Midjourney shows you pairs of images and asks which you prefer. The pattern of your choices trains a profile that is applied to your generations when you use --p. A strong profile requires 500 or more rankings.
Moodboard
A collection of saved images in a Midjourney project used as visual reference for future generations. Images saved to a moodboard can be used as style references to anchor new generations to an established aesthetic.
General AI image generation terms
Upscaling
Increasing the resolution of a generated image. In Midjourney this means selecting a specific image from the generation grid (U1 through U4) to produce the full-resolution version. In other tools upscaling may mean using a separate AI upscaling tool to increase resolution beyond what the generator originally produced.
Inpainting
Editing a specific area of an existing image while leaving the rest unchanged. You select a region of the image and describe what should replace it. Midjourney calls this the Edit feature. Useful for fixing specific problems in otherwise good images - replacing an unwanted element, fixing distorted hands, correcting a label.
Outpainting
Extending an image beyond its original borders. The AI generates new content that continues the image naturally. Useful for changing aspect ratios, adding context to a tight crop, or extending a background.
Text to image
The basic AI image generation workflow. You write a text description and the AI generates an image matching that description. All AI image generators support text to image as their core function.
Image to image
Using an existing image as a reference or starting point for generation. You upload an image and the AI generates a new image influenced by or based on it. The strength of the image influence is usually controllable.
Image to video
Using a still image as the starting point for video generation. The AI animates the still image by adding motion. This is the recommended workflow for AI video production - generate a strong still image first then animate it rather than generating video from text alone.
Text to video
Generating video directly from a text description without a reference image. Sora is the strongest text to video tool currently available. Results are more variable than image to video workflows but improving rapidly.
Diffusion model
The underlying technology that most AI image generators use. Diffusion models learn to generate images by learning to reverse a process of adding noise to images. They start with random noise and gradually refine it into a coherent image guided by your prompt. Midjourney, Stable Diffusion, and DALL-E are all diffusion models.
Latent space
The mathematical representation of images that a diffusion model works in. When you generate an image the model navigates through latent space guided by your prompt to find an image that matches your description. The seed number determines the starting point in latent space for each generation.
CFG scale (classifier-free guidance scale)
A parameter in Stable Diffusion and some other tools that controls how closely the generation follows your prompt versus how much creative freedom the model takes. High CFG scale produces outputs that closely match your prompt but can look oversaturated or artificial. Low CFG scale gives the model more creative freedom. Midjourney's --stylize parameter serves a similar function.
LoRA (low-rank adaptation)
A technique for fine-tuning AI models on specific styles, subjects, or aesthetics without retraining the entire model. LoRAs are common in Stable Diffusion workflows and allow users to train the model on their specific brand aesthetic, product appearance, or artistic style. Not directly applicable to Midjourney but important to understand in the broader AI image landscape.
Checkpoint
A saved state of a trained AI model. In Stable Diffusion different checkpoints produce dramatically different aesthetic qualities. Choosing the right checkpoint for your use case is as important as writing a good prompt. Not applicable to Midjourney which manages model versions through the --v parameter.
Sampling method
The algorithm used by a diffusion model to move from noise to image. Different sampling methods produce different results and have different speed and quality trade-offs. Relevant in Stable Diffusion. Not directly controllable in Midjourney.
Resolution
The pixel dimensions of the generated image. Higher resolution images contain more detail and are suitable for larger print sizes. AI generators have default output resolutions that can sometimes be increased through upscaling.
Aspect ratio
The proportional relationship between image width and height. 1:1 is square. 16:9 is wide landscape. 9:16 is tall portrait. Setting the correct aspect ratio for your intended platform before generating prevents awkward cropping later.
Video generation terms
Motion prompt
The text description used specifically for AI video generation. Different from an image prompt because it must describe what moves, how the camera moves, and what happens over time - not just what the scene looks like.
Camera preset
A named camera movement configuration in video generation tools, particularly Higgsfield. Presets like Slow Dolly In, Crane Up, FPV, and Orbit apply specific professional camera movements to your generation without requiring you to describe the mechanics of the movement in technical detail.
Motion intensity
The amount of movement in an AI video generation. Low motion intensity produces subtle atmospheric movement - rising steam, floating particles, barely perceptible camera drift. High motion intensity produces dynamic energetic movement with clear camera action and strong subject motion.
Temporal consistency
The degree to which elements in a video remain consistent from frame to frame. High temporal consistency means the subject looks the same throughout the clip. Low temporal consistency produces morphing, flickering, or distortion over time. One of the key quality metrics for evaluating AI video output.
FPS (frames per second)
The number of individual frames in each second of video. Standard video is 24fps. Smooth video is 30fps or 60fps. AI video generators have fixed output frame rates - Higgsfield outputs at 30fps, Midjourney Video at 24fps.
Loop
A video clip where the last frame seamlessly transitions back to the first frame creating an infinitely repeating video. Seamless loops are essential for website hero videos and animated ads. Most AI video tools have a loop mode or loop generation option.
Quality and output terms
Artifact
An unwanted visual error in an AI-generated image. Common artifacts include extra fingers, distorted facial features, blurry patches, impossible geometry, double edges, and texture inconsistencies. Artifacts are one of the primary quality issues to evaluate when selecting images for professional use.
Hallucination
When an AI generates something that was not in your prompt and does not belong in the scene. A product image that spontaneously generates text on the label that was never there. A person who appears in a scene you described as empty. Hallucinations are reduced through more specific prompting and negative prompts.
Coherence
How logically consistent and internally correct a generated image is. A coherent image has lighting that makes physical sense, shadows in the right directions, materials that behave correctly, and subjects that look anatomically correct. Incoherence is one of the most common quality issues in AI generation.
Photorealism
The quality of an AI image that makes it appear to be a real photograph rather than a generated image. Achieving high photorealism requires specific prompt language - camera and lens specifications, film stock references, natural imperfections, and physical lighting descriptions.
Bokeh
The aesthetic quality of the out-of-focus areas in an image. Professional photography uses shallow depth of field to create smooth, creamy background blur (bokeh) that isolates the subject. In AI prompts specifying a wide aperture (f/1.4, f/2.0) and a telephoto lens (85mm, 100mm) produces natural-looking bokeh.
Depth of field
The range of distance in an image that appears acceptably sharp. Shallow depth of field means only a narrow band of distance is in focus - the subject is sharp but the background is blurred. Deep depth of field means everything from foreground to background is sharp. Controlled in AI prompts through aperture specification.
Workflow terms
Batch generation
Running multiple related prompts in a single focused session rather than one at a time across different days. Batching produces more consistent results because the same parameters and creative context apply across all generations in the session.
Prompt template
A reusable prompt structure with fixed brand elements and variable content slots. Templates are the foundation of efficient AI content production. You build a template once for each content type and use it for every piece of that type rather than writing fresh prompts each time.
Content library
A organized collection of generated images ready for use. A healthy content library contains more images than immediately needed - a buffer of on-brand content that can be drawn from quickly when needed without requiring a new generation session.
Style transfer
Applying the visual aesthetic of one image to another. In Midjourney this is done with --sref. The style - lighting quality, color palette, texture, mood - transfers from the reference image to the new generation while the content is determined by the prompt.
Now put the vocabulary to work
Understanding the terminology is the first step. Using it is the second.
Our free AI Image Prompt Generator lets you put this vocabulary into practice immediately. Describe what you want to create in plain English, select your content type, mood, and style, and get 5 ready-to-use prompt variations - safe, cinematic, editorial, minimal, and experimental - each built using the full 6-layer framework.