Plain definitions of 60 terms you meet when writing prompts for AI image and video models, from aspect ratio and seed to camera movement and LoRA. Each entry links to related terms and to a Genvizu page where you can see the idea in practice.
Adaptive ratio is an aspect-ratio option in which the output frame follows the reference material instead of a fixed shape. It avoids forcing a crop when your reference is unusual. The trade-off is that you do not fully control the final frame size, so choose a fixed ratio when the delivery format is strict.
Related: Aspect ratio · Reference image · Video-to-video
On Genvizu: /model/seedance-2-5
Aspect ratio is the proportion of an image or video's width to its height, written as two numbers such as 16:9 or 9:16. It decides the framing before any detail is generated: wide ratios suit landscapes and film, tall ratios suit phone screens and posters. Each model supports its own list of ratios, so the available choices differ from model to model.
Related: Resolution · Outpainting · Composition
On Genvizu: /model/gpt-image-2.5 · /category/video
Audio-to-video is a generation mode in which a sound track, such as speech or music, guides the video that is produced, so motion or mouth movement follows the audio. Some models accept reference audio clips alongside text and images. Support, clip limits and how closely the picture follows the sound depend on the model.
Related: Lip sync · Image-to-video · Text-to-video
On Genvizu: /model/seedance-2-5 · /category/video
Camera movement is the way a virtual camera moves during a video shot, described in prompts with terms such as pan, tilt, dolly, zoom and orbit. A pan turns the camera sideways, a tilt turns it up or down, a dolly moves it forward or back, a zoom changes focal length, and an orbit circles the subject. Naming one movement per shot gives the most predictable results.
Related: Shot size · Keyframe · 360° turnaround
On Genvizu: /blog/ai-video-camera-movement-prompts · /ai-video-prompt-generator
CFG, short for classifier-free guidance, is a setting in diffusion models that controls how strongly the output is pushed to follow the prompt. Low values give the model more freedom and often softer, more varied results; high values follow the wording more literally but can look over-processed. Many hosted models fix this value and do not expose it.
Related: Steps · Seed · Diffusion model
On Genvizu: /ai-prompt-generator
Chaos is a Midjourney parameter that controls how different the images in one result set are from each other. A low value yields four similar interpretations of the prompt, while a high value yields more unexpected compositions and styles. It is useful early in exploration and less useful once you know what you want.
Related: Stylize · Prompt weight · Tile
On Genvizu: /model/midjourney-8-2 · /midjourney-prompt-helper
Character consistency is the ability to keep the same person, creature or mascot recognisable across several generated images or shots. Common methods are supplying reference images, repeating a precise character description, and fixing style words. It matters most for storyboards, comics and brand mascots, where small drifts in face or clothing break continuity.
Related: Reference image · Multi-image reference · Storyboard
On Genvizu: /blog/ai-character-consistency-prompts · /model/nano-banana-2.1
A checkpoint is a saved set of trained model weights, so the same architecture can have many checkpoints with different strengths such as photorealism or illustration. In open-source image generation, people swap checkpoints to change the base look. Hosted services usually hide this choice and present finished models instead.
Related: LoRA · Diffusion model · Latent space
On Genvizu: /explore
Color grading is the deliberate adjustment of colour, contrast and tone across an image or video to set mood and keep a consistent look. In prompts it appears as phrases such as teal and orange, faded film or high-contrast monochrome. It is separate from lighting: lighting shapes what the camera sees, while grading shapes how that footage is finished.
Related: Lighting terms · Film emulation · Composition
On Genvizu: /tools/film-emulator · /blog/monocolor-editorial-print-skill
Commercial use means using generated images or videos in paid or business contexts such as ads, products or client work. Whether it is allowed depends on the model provider's terms and the platform's own rules, and these can differ per model. Genvizu flags some models, including its Midjourney models, as commercially restricted, so read the terms shown for the model you pick.
Related: Watermark · Content moderation · Prompt gallery
On Genvizu: /model/midjourney-8-2 · /docs/model-pricing
Composition is how elements are arranged inside the frame, including placement, balance, leading lines and negative space. In prompts it is controlled with words like centred, symmetrical, rule of thirds, bird's-eye view or close crop. Strong composition wording is often the quickest way to make an image look planned instead of accidental.
Related: Rule of thirds · Depth of field · Aspect ratio
On Genvizu: /blog/ai-poster-prompt-design · /category/ads-product
Content moderation is the set of automated and manual checks a service applies to prompts and outputs to block disallowed material. A prompt can be refused even when your intent is harmless, because filters work on wording and context. If a generation is refused, rewording the request more plainly is usually the first thing to try.
Related: Prompt injection (image misuse) · Commercial use / license · Watermark
On Genvizu: /docs/user-guide
Depth of field is the range of distance in a scene that appears sharp. A shallow depth of field keeps the subject crisp and blurs the background, often called bokeh, while a deep one keeps near and far both sharp. In prompts, phrases like f/1.8, shallow focus or tilt-shift are common ways to steer it.
Related: Composition · Lighting terms · Shot size
On Genvizu: /category/portrait
A diffusion model is a type of generative model that learns to create images or video by gradually removing noise from random static, guided by a text prompt. Most current image and video generators are built on this idea. The number of denoising steps and the guidance strength affect speed and how closely the result follows the prompt.
Related: Latent space · Steps · CFG / guidance scale
On Genvizu: /explore
Duration is the length of a generated video clip, usually set in whole seconds. Each model offers its own range: on Genvizu it runs from 1-second clips on Grok Imagine Video 1.5 to 30 seconds on Seedance 2.5, while Veo 3.1 offers fixed 8-second clips. Longer clips are harder to keep consistent, so many projects join several short shots.
Related: Frame rate · Text-to-video · Keyframe
On Genvizu: /model/seedance-2-5 · /model/veo-3.1
Film emulation is the process of making digital footage or images look like a particular analogue film stock, with its characteristic colour response, contrast and grain. It can be done through grading presets or by naming a film look in the prompt. It sits between colour grading and style reference, since it borrows a recognisable photographic finish.
Related: Color grading · Style reference · Lighting terms
On Genvizu: /tools/film-emulator
First and last frame control means giving a video model a starting image and an ending image so it generates the motion that connects them. It is used for transitions, product reveals and loops. Support varies: some models take two frames natively, while others need the stills placed first among the references and named in the prompt.
Related: Keyframe · Image-to-video · Reference image
On Genvizu: /model/gemini-omni-flash-1-1 · /blog/canvas-video-transitions-ai-seamlessly-connects-clips
Frame rate is how many individual frames are shown per second of video, written as fps. Common values are 24 for a cinematic feel, 30 for general web video and 60 for smooth motion. Most hosted generators fix the frame rate for you, so duration in seconds, not frame count, is what you choose.
Related: Duration · Keyframe · Motion strength
On Genvizu: /category/video
Image-to-image is a generation mode that starts from a supplied picture and a prompt, producing a new image that keeps some of the original's structure or content while changing style, details or setting. It is used for restyling, variations and edits. How much of the original survives depends on the model and its settings.
Related: Reference image · Inpainting · Text-to-image
On Genvizu: /model/nano-banana-2.1 · /image-to-prompt
Image-to-prompt, also called prompt reversal, turns an existing picture into descriptive text that can be edited and reused as a prompt. It helps you learn what language describes a look and rebuild a similar image. The text it returns is a starting point, since it describes what is visible, not the original settings.
Related: Prompt · Reference image · Prompt gallery
On Genvizu: /image-to-prompt
Image-to-video is a generation mode in which a still image becomes the starting point of a video clip, and the prompt describes the motion that follows. It gives more control over the look of the first frame than text-to-video. Results depend on how clearly the prompt describes camera and subject movement.
Related: Text-to-video · First / last frame · Camera movement
On Genvizu: /blog/seedance-2-0-image-to-video-prompt · /category/video
Inpainting is editing a chosen region of an existing image, usually marked with a mask, while leaving the rest untouched. You describe what should appear in the masked area, such as a different object or a repaired background. Some models perform similar local edits from instructions alone, without a drawn mask.
Related: Outpainting · Image-to-image · Upscaling
On Genvizu: /model/gpt-image-2.5-sunburst
A keyframe is a frame that defines an important point in a motion, such as the start, the end or a pose in between. In AI video, keyframes are stills you supply or generate to anchor a sequence, and the model fills in motion between them. Planning keyframes first makes multi-shot work easier to keep consistent.
Related: First / last frame · Storyboard · Image-to-video
On Genvizu: /blog/ai-video-storyboarding-timeline-15s-shorts
Latent space is the compressed internal representation in which many generative models work, instead of operating on raw pixels. Images are encoded into this smaller space, changed there, then decoded back into a picture. It makes generation faster and is why similar prompts can land near each other in style.
Related: Diffusion model · Checkpoint · Seed
On Genvizu: /explore
Lighting terms are words that describe how a scene is lit, and they strongly shape mood in generated images. Rim light outlines a subject's edge from behind, a softbox gives broad soft light, and golden hour means the warm low light shortly after sunrise or before sunset. Naming the light source and direction works better than just saying good lighting.
Related: Color grading · Depth of field · Product shot
On Genvizu: /tools/relight · /category/portrait
Lip sync is matching a character's mouth movements to spoken audio so speech looks natural on screen. In AI video it can be produced by models that take an audio track as input, or by a separate post-processing step on finished footage. Accuracy varies with the model, language and how much of the face is visible.
Related: Audio-to-video · Image-to-video · Text-to-video
On Genvizu: /category/video
LoRA, short for low-rank adaptation, is a small add-on file that adjusts a base image model toward a particular style, character or subject without retraining the whole model. LoRAs are common in open-source workflows and can be combined with a checkpoint. Hosted platforms usually offer finished models rather than loadable LoRAs.
Related: Checkpoint · Diffusion model · Style reference
On Genvizu: /explore
A mockup is a realistic image that shows a design applied to an object or setting, such as a logo on packaging, a poster on a wall or an app on a phone. Generated mockups are used to present concepts before production. Clear descriptions of the object, material and surroundings matter more than adjectives about quality.
Related: Product shot · Typography prompt · Composition
On Genvizu: /category/brand-logo · /blog/ai-packaging-design-prompts
Motion strength is a setting or prompt cue that controls how much movement appears in a generated video. Low strength keeps the scene nearly still, with gentle drift, while high strength allows large subject and camera motion but risks distortion. Where no slider exists, words such as slow, subtle or fast in the prompt play the same role.
Related: Camera movement · Frame rate · Image-to-video
On Genvizu: /category/video
Multi-image reference means supplying several pictures at once to guide one generation, for example a face, a product and a background. Limits differ by model: on Genvizu, GPT Image 2.5 accepts up to 16 reference images, Nano Banana 2.1 up to 10 and Seedream 5.0 Pro up to 5. Labelling each image in the prompt helps the model use it correctly.
Related: Reference image · Character consistency · Image-to-image
On Genvizu: /model/gpt-image-2.5 · /model/nano-banana-2.1
A negative prompt is a second text field that lists what the model should avoid, such as blurry, extra fingers or watermark. It is common in open-source diffusion tools. Many hosted models have no separate field, so the usual alternative is to state the wanted result positively and describe constraints inside the main prompt.
Related: Prompt · CFG / guidance scale · Watermark
On Genvizu: /ai-prompt-generator
Outpainting extends an image beyond its original borders by generating new content that continues the scene. It is used to change aspect ratio, add space for text or reveal more of a setting. Results are best when the edge of the original has clear context, such as a horizon or a pattern that the model can continue.
Related: Inpainting · Aspect ratio · Image-to-image
On Genvizu: /model/gpt-image-2.5
A product shot is an image that presents a product clearly for selling or advertising, typically on a clean background or in a lifestyle setting. Prompts name the object, material, angle, lighting and surface. Marketplaces often require plain backgrounds, while lifestyle shots show use in context.
Related: Mockup · Lighting terms · 360° turnaround
On Genvizu: /category/ads-product · /blog/ai-product-scene-tutorial
A prompt is the text instruction you give a generative model to describe what to create. Strong prompts name the subject, setting, style, lighting, camera and mood, in plain and specific language. Length limits differ by model, from about a thousand characters on Midjourney 8.2 to twenty thousand on GPT Image 2.5.
Related: Negative prompt · Prompt length / tokenizer · Prompt gallery
On Genvizu: /ai-prompt-generator · /model/midjourney-8-2
A prompt gallery is a browsable collection of generated works published together with the prompts and settings that produced them. It lets you study what wording gives which result, then reuse an idea as a starting point. On Genvizu you can filter works by model, category or tag and load one straight into the generator.
Related: Prompt · Skills · Workflow / canvas
On Genvizu: /explore · /category/illustration-3d
Prompt injection is an attempt to slip instructions into an input so the system behaves in a way its owner did not intend. For image tools, the related misuse is wording designed to get around content filters, or text hidden inside a reference image. Services defend against it with moderation checks, so such attempts are usually refused.
Related: Content moderation · Prompt · Commercial use / license
On Genvizu: /docs/user-guide
Prompt length is how much text a model accepts, counted in characters or tokens, where a tokenizer splits text into small units the model reads. Limits vary widely: on Genvizu, Midjourney 8.2 allows 1,024 characters, Seedream 5.0 Pro 3,000 and GPT Image 2.5 20,000. Very long prompts can dilute focus, so lead with the most important details.
Related: Prompt · Prompt weight · Text rendering in images
On Genvizu: /model/midjourney-8-2 · /model/gpt-image-2.5
Prompt weight is a way to make some words count more than others in a prompt. Midjourney does this with a double colon and a number after a phrase, while other tools use brackets or numeric weights. Support depends on the model, so check the syntax guide for the one you use before relying on it.
Related: Stylize · Chaos · Prompt
On Genvizu: /midjourney-prompt-helper · /blog/midjourney-8-2-prompt-guide
A reference image is a picture you supply to guide a generation, for a subject, style, composition or colour palette. It reduces guessing compared with text alone. Models differ in how many references they accept and how closely they follow them, and some, like Grok Image 2 on Genvizu, accept none.
Related: Multi-image reference · Style reference · Image-to-image
On Genvizu: /model/gpt-image-2.5 · /image-to-prompt
Resolution is the pixel size of the output, often named by tier such as 1K, 2K and 4K for images or 480p, 720p and 1080p for video. Higher tiers carry more detail but take longer. Available tiers depend on the model: on Genvizu, GPT Image 2.5 offers 1K, 2K and 4K, while Seedance 2.5 offers 480p and 720p.
Related: Aspect ratio · Upscaling · Frame rate
On Genvizu: /model/gpt-image-2.5 · /model/seedance-2-5
The rule of thirds is a composition guideline that divides the frame into a three-by-three grid and places key subjects along the lines or at their intersections. It tends to feel more balanced than centring everything. In prompts, phrases like subject on the left third create a clear space for text or negative space on the other side.
Related: Composition · Depth of field · Shot size
On Genvizu: /blog/ai-poster-prompt-design
A sampler is the algorithm that carries out the denoising steps in a diffusion model, deciding how noise is removed at each stage. Different samplers trade speed, sharpness and variety, and are named things like Euler or DPM. Most hosted services pick the sampler for you, so it is mainly a concern in open-source tools.
Related: Steps · CFG / guidance scale · Seed
On Genvizu: /explore
A seed is the number that initialises the random noise a generation starts from. With the same seed, prompt and settings, a model that exposes it can reproduce nearly the same image, which makes controlled comparisons possible. Many hosted models do not expose the seed, so exact repeats are not always available.
Related: Steps · CFG / guidance scale · Diffusion model
On Genvizu: /ai-prompt-generator
Series consistency is keeping a shared look, palette and character design across many generated pieces, such as a set of covers or a sticker pack. Reusing the same model, style words and references is the main method. Batch workflows and packaged Skills help because they fix those inputs once.
Related: Character consistency · Style reference · Skills
On Genvizu: /tools/cover-maker · /blog/monocolor-editorial-print-skill
Shot size describes how much of a subject fills the frame, from an extreme close-up of a detail to a wide shot that shows the whole setting. Common sizes are close-up, medium shot and wide shot. Naming the shot size in a prompt is a simple way to control framing, especially for video where each shot needs its own distance.
Related: Camera movement · Composition · Storyboard
On Genvizu: /blog/ai-video-camera-movement-prompts
Skills are packaged, parameterised prompt recipes that turn a short brief into a structured prompt aimed at a suitable model. Genvizu's Skills cover tasks such as product photo sets, cover design and sticker packs. They save you from writing long prompts from scratch and keep outputs consistent across a batch.
Related: Prompt · Workflow / canvas · Prompt gallery
On Genvizu: /canvas · /blog/monocolor-editorial-print-skill
Steps is the number of denoising iterations a diffusion model runs to produce an image. More steps can add refinement but with diminishing returns and longer waits, while very few steps give rough results. Some fast models are tuned to work in very few steps, and hosted services often set this automatically.
Related: Sampler · CFG / guidance scale · Diffusion model
On Genvizu: /model/z-image
A storyboard is a sequence of panels or shots that plans a video or story before it is produced, showing framing, action and order. With AI tools, each panel can be generated as an image, then turned into a clip. Planning the cuts first matters because most video models generate one continuous shot at a time.
Related: Keyframe · Shot size · Character consistency
On Genvizu: /blog/video-to-storyboard-sheet-genvizu-infinite-canvas · /blog/ai-video-storyboarding-timeline-15s-shorts
A style reference is an image or code supplied to carry a look, such as colour, texture and brushwork, into a new generation without copying the subject. It differs from a content reference, which carries what is shown. It is useful for keeping a series visually consistent across different subjects.
Related: Reference image · Character consistency · Stylize
On Genvizu: /blog/monocolor-editorial-print-skill
Stylize is a Midjourney parameter that sets how strongly the model applies its own aesthetic taste versus following your prompt literally. A low value sticks closer to the wording, while a high value yields more artistic, polished and sometimes less literal images. It is worth adjusting when output feels either too plain or too stylised.
Related: Chaos · Prompt weight · Style reference
On Genvizu: /model/midjourney-8-2 · /midjourney-prompt-helper
Text rendering is a model's ability to draw readable, correctly spelled words inside an image, such as signs, posters or labels. It is a known weak point for many generators and an advertised strength of others. For best results, put the exact wording in quotation marks, keep it short and state where it should appear.
Related: Typography prompt · Prompt length / tokenizer · Composition
On Genvizu: /model/gpt-image-2 · /model/qwen-image-2.1
Text-to-image is a generation mode that creates a new image from a written prompt alone, with no starting picture. It is the most common way to use an image model. Specific wording about subject, style, lighting and composition gives more predictable results than a short or abstract phrase.
Related: Image-to-image · Prompt · Aspect ratio
On Genvizu: /ai-prompt-generator
Text-to-video is a generation mode that creates a video clip from a written prompt alone. A good prompt describes the subject, the action, the setting and one camera movement. Clip length, resolution and whether sound is generated depend on the model, so check the model page before planning a shot.
Related: Image-to-video · Camera movement · Duration
On Genvizu: /blog/seedance-2-0-text-to-video-prompt · /ai-video-prompt-generator
Tile is a Midjourney parameter that makes the generated image repeat seamlessly when placed side by side. It is used for patterns, textures and wallpaper. Without it, edges rarely match, so a repeat looks stitched together. Check the result by placing copies next to each other before using it.
Related: Chaos · Stylize · Mockup
On Genvizu: /model/midjourney-8-2
A typography prompt is a prompt written to produce designed lettering, such as a poster headline or a logo wordmark. It specifies the exact text, the typeface style, hierarchy, colour and where the words sit. Pairing a model with strong text rendering and a short exact string gives the cleanest result.
Related: Text rendering in images · Mockup · Composition
On Genvizu: /blog/ai-typography-poster-prompt-guide · /category/brand-logo
Upscaling increases the resolution of an image or video after it has been generated, adding detail so it holds up at larger sizes. It is done with a separate model rather than by regenerating. Genvizu offers video upscaling on the Topaz engine. Upscaling cannot fix a weak composition, only sharpen what exists.
Related: Resolution · Inpainting · Frame rate
On Genvizu: /blog/ai-video-upscaling-topaz-engine
Video-to-video is a generation mode that takes an existing clip and a prompt and produces a new clip that keeps its motion or structure while changing style or content. Some models accept reference videos alongside text and images. Limits on the number and total length of reference clips depend on the model.
Related: Text-to-video · Image-to-video · Motion strength
On Genvizu: /model/seedance-2-pro · /blog/video-to-script-ai-tool
A watermark is a visible or invisible mark added to an image or video to show its origin or that it is AI-generated. Some providers embed invisible watermarks such as SynthID, and some add a visible label. Removing or hiding such marks to mislead others can break platform terms, so check the rules for your use.
Related: Commercial use / license · Content moderation · Negative prompt
On Genvizu: /docs/model-pricing
A workflow or canvas is a node-based workspace where prompts, reference images and generations are connected as a graph, so a multi-step idea stays visible instead of scrolling away in a chat. Genvizu's Canvas mode links prompt nodes, references, generations and agent steps, and results can be reused as inputs to the next step.
Related: Skills · Storyboard · Prompt gallery
On Genvizu: /canvas · /blog/genvizu-infinite-canvas-complex-workflows
A 360-degree turnaround shows a subject from all sides, either as a set of views or as a rotating clip, so you can judge its form from every angle. It is used for characters, products and design review. Generating consistent views is hard, so a fixed background and steady lighting help.
Related: Product shot · Camera movement · Character consistency
On Genvizu: /blog/360-degree-ai-image-prompts
