The words, defined

Text-to-video

Text-to-video is the generation of a video clip from a written prompt alone. The model synthesizes subject, motion, camera and lighting from the description, typically producing clips of a few seconds.

Image-to-video

Image-to-video animates a still image into a video clip: the picture fixes the subject and framing, and a prompt describes the motion. It is the standard way to keep a specific face or product consistent in generated video.

Camera move

A camera move is how the virtual camera travels during a clip — pan, push-in, tracking, orbit, crane, dolly zoom. In AI video it is prompted, not rigged, and it shapes the feel of a shot more than the subject does.

Dolly zoom

A dolly zoom moves the camera toward or away from the subject while zooming the opposite way, so the subject stays the same size while the background stretches or compresses. It is the "vertigo effect" from Hitchcock and Jaws.

Aspect ratio

Aspect ratio is the width-to-height proportion of a frame: 16:9 for widescreen and YouTube, 9:16 for vertical shorts, 1:1 for square feeds, 21:9 for cinema. In AI generation it is a parameter on modern engines and a prompt hint on older ones.

Storyboard

A storyboard is a film planned shot by shot — each scene's framing, action and length in order. In AI video it is the way past the clip-length limit: several clips with one look, played as one cut.

Take

A take is one rendered attempt at a scene. Rendering several takes of the same prompt and keeping the best is how generated video gets directed rather than accepted.

First/last frame

First/last-frame generation gives a video engine both a starting image and an ending image; the model interpolates the motion between them. It is the most precise way to control where a clip ends.

Tempo

Tempo is the pacing of motion within a clip — calm, dreamy, dynamic, tense or chaotic. It shapes how fast subjects and camera move and how much energy a shot carries.

Anamorphic lens

An anamorphic lens squeezes a wide image onto the sensor and is unsqueezed in post, producing the cinematic signature of oval bokeh, horizontal lens flares and a gentle widescreen stretch.

Upscaling

Upscaling enlarges a generated image or clip after the fact, inventing the extra pixels with a second model. It is how a 720p render becomes a 1080p or 4K deliverable without being regenerated.

Lip-sync

Lip-sync animates a face so the mouth matches an audio track you supply. It is what turns a still portrait or a silent clip into someone appearing to speak your words.

B-roll

B-roll is the supporting footage cut over narration or between the main shots — hands, details, a street, the product on a table. It carries the edit while the voice carries the argument.

Establishing shot

An establishing shot is the wide frame that tells the audience where they are before the scene starts. It is usually the first shot of a sequence and often the only one that shows the whole location.

Keyframe

A keyframe is a frame whose content is fixed rather than generated, and which the motion is built around. In AI video the start frame and, where supported, the end frame are the keyframes.

Reference image

A reference image is a picture supplied alongside a prompt so the model holds something from it — a face, a product, a style — across new renders. It is the basis of character consistency in AI image and video.

Film grain

Film grain is the fine, random texture of photochemical film, added to digital renders to suggest analog capture and to hide the too-clean look of synthesis.

Image-to-image

Image-to-image edits or restyles an existing picture guided by a prompt, keeping its composition while changing details, style or subject. Reference-guided generation is its most common creative form.

Contre-jour

Contre-jour is backlighting: the light source sits behind the subject, producing a rim-light halo, silhouettes and lens glow. It is the classic golden-hour look.

Inpainting

Inpainting regenerates a selected region of an image while leaving the rest untouched — masking a hand and redrawing it, removing an object, changing one detail. The mask is the instruction.

Voice cloning

Voice cloning creates a synthetic voice from a recording so new text can be spoken in that voice. Modern engines need about a minute of clean speech and reproduce timbre and cadence across languages.

Section tags

Section tags are markers like [Verse], [Chorus] and [Bridge] written on their own lines in song lyrics, telling an AI music generator where each part of the song begins. They shape the arrangement: a chorus that returns, a bridge that changes the mood, an outro that lands the ending.

Text-to-speech (TTS)

Text-to-speech converts written text into spoken audio. Expressive TTS engines read punctuation as performance and expose controls like speed and expressiveness, producing natural narration and dialogue.

Speech-to-text (STT)

Speech-to-text transcribes spoken audio into written text. Paired with a chat model and TTS it forms a voice conversation loop.

Voice-over

A voice-over is narration laid over picture rather than spoken on camera. It carries the argument of an explainer, an ad or a documentary while the footage illustrates it.

Diarization

Diarization is the part of transcription that works out who spoke when, labelling a transcript by speaker. It is what turns a wall of text from an interview into a readable two-column conversation.

Negative prompt

A negative prompt lists what the model should avoid — text, watermarks, extra fingers, blur. It steers generation away from those features instead of hoping the main prompt implies their absence.

CFG scale

CFG scale (classifier-free guidance) sets how strictly a diffusion model follows the prompt. Low values give looser, more varied images; high values follow the words literally at the cost of naturalness.

Seed

A seed is the number that initializes a generation's randomness. The same prompt, settings and seed reproduce the same output, which turns re-rendering into revision instead of a gamble.

Prompt enhancer

A prompt enhancer rewrites a rough idea into a prompt structured the way a generation engine responds to — subject, motion, camera, light, mood — or invents one from scratch.

Director's rack

The director's rack is eroq's set of look controls — film type, era, tempo, camera gear, lens, aperture, palette, lighting — folded into every prompt server-side so a whole film shares one aesthetic.

Depth of field

Depth of field is how much of a shot is in focus from front to back. A shallow depth of field isolates the subject against a soft background; a deep one keeps the whole scene sharp.

Color palette

A color palette is the restricted set of hues a piece is graded toward — neon noir, warm film, bleach bypass. Committing to one is what makes separate shots look like one piece of work.

UGC ad

A UGC ad (user-generated-content ad) is a video ad that looks filmed by a customer or a creator, usually on a phone and talking straight to the camera about the product. It works because it reads as a person's recommendation rather than a brand's message.

Ad creative

The ad creative is the part of an ad people see and hear: the picture or the video, the headline, the text and the call to action. Targeting decides who sees an ad; the creative decides whether they stop and act.

Creative fatigue

Creative fatigue is the decline of an ad's results as the same audience sees it again and again: fewer people stop, fewer click, and each result costs more. It is the main reason ad creatives have to be refreshed.

Hook

The hook is the first second or two of an ad: the image, line or question that makes someone stop scrolling. If the hook fails, nothing after it is seen.

Call to action (CTA)

The call to action tells the viewer what to do next: shop, try, sign up, learn more. In ad platforms it is both a line in the copy and a button with a fixed label.

Brand kit

A brand kit is the set of rules every piece of a brand's content follows: name, logo, colors, fonts, tone of voice, and what it sells to whom. It keeps hundreds of ads recognizably from the same brand.

Selling angle

A selling angle is the single reason an ad gives to buy: the offer, the main benefit, an ingredient, the way it is used, social proof, a problem it solves. Each ad should stand on one angle, so a test shows which reason sells.

Ad variations

Ad variations are versions of one ad that change a single element (the face, the setting, the headline, the format) while the idea stays. They let a winning ad keep running longer and show which details matter.

Ad placement

An ad placement is where an ad is shown: a Facebook or Instagram feed, Stories and Reels, TikTok's feed, a pin, a banner on a website. Each placement has its own picture size and shows a set amount of text.

Uncensored AI

Uncensored AI describes models that render what the prompt asks for — dark, violent or provocative fiction included — instead of refusing it, governed by a written acceptable-use policy rather than by a probabilistic content filter. It has never meant "no rules".

Private mode

Private mode is a setting that returns an AI render to you without keeping it: no stored file, no library entry, and the prompt starred out of the usage records. It is how a creative tool lets you work on something without leaving a copy on its side.

Credits

Credits are a prepaid unit for AI generation: each call costs a published number of credits instead of a token bill, which makes budgets predictable. On eroq, 1 credit is about $0.01 at the entry pack.

MCP (Model Context Protocol)

MCP is an open protocol that lets AI assistants call external tools. An MCP server exposes capabilities — like generating a video — that ChatGPT, Claude or a coding agent can invoke in conversation.

SSE streaming

Server-sent events stream a chat reply token by token over one HTTP connection, so users see text as it is generated instead of waiting for the whole completion.

Element

On eroq, an Element is a reusable place, object or style saved once and @mentioned in prompts, so the same location, product or look appears across renders and across a team.

Character (persistent)

A persistent character is an AI cast member defined once — face references, persona, voice — and reused across images, video and dialogue so they stay the same person in every scene.

Remix

Remix reloads a finished render's complete recipe — prompt, model, camera, rack settings, length — into the composer so it can be revised rather than rebuilt from memory.

Webhook

A webhook is a signed HTTP callback a platform sends to your server when an asynchronous job finishes, so you do not have to poll. Video generation, which takes minutes, is the canonical case.

Workspace

A workspace is a shared account space with members, seats and one credit wallet, so a team spends and creates together while each person keeps their own login.

Watermark

A watermark is a visible mark burned into generated media, usually a logo in a corner, identifying the tool that made it. Many generators apply one on free or lower tiers and remove it higher up.

Rate limit

A rate limit caps how many requests an account may send per minute. It protects shared capacity and is the reason a burst of parallel calls can fail while the same calls spread out succeed.

API key

An API key is the secret string that authenticates a program against an API, sent as a bearer token on every request. It identifies the account paying for the call.