Giải thích nhiệm vụ thế hệ AI

A character reference is one or more images supplied alongside a prompt so a model reproduces that specific person, creature or object instead of inventing a new one. It conditions identity rather than composition, which is what makes character consistency possible across separate generations.

subject reference · character consistency · cref · sref · reference image

A depth map is a grayscale image where each pixel encodes distance from the camera instead of color, conventionally with near surfaces bright and far ones dark. Generation pipelines use a depth map as a structural control signal, because it carries the layout and volume of a scene without carrying its style, so you can change how an image looks while keeping where everything is.

depth estimation · monocular depth · depth pass · z-depth

Frame interpolation synthesizes new frames between existing ones, either to raise a clip's frame rate or to stretch it into slow motion. Modern methods estimate how every pixel moves between two frames and render the in-between along that motion, rather than cross-fading the two images.

video frame interpolation · keyframe interpolation · interpolation · frame blending · fps boost

A green screen is an evenly lit backdrop in a colour that does not occur in skin or hair, so software can single that colour out and make it transparent. Removing it is called chroma key, or keying, and the result is an alpha channel that lets the subject be composited over anything else.

chroma key · keying · blue screen · chroma keying · alpha channel

Image segmentation divides a picture into labeled regions at the pixel level, producing a mask rather than a bounding box. Inside a generation pipeline it is the step that decides which pixels an edit is allowed to touch, which puts it underneath background removal, object removal, masked inpainting, and any per-subject adjustment.

semantic segmentation · instance segmentation · masking · cutout

Image to image, usually shortened to i2i, is the generation task where an input image conditions the output. The model rewrites an existing picture instead of starting from noise alone, so composition, color, and pose carry over to whatever degree the strength setting allows.

img2img · image to image translation · what is image to image · reference image

Image to video, usually shortened to i2v, is the generation task where a still image becomes the starting frame of a clip and the model only has to invent motion. Appearance is locked by the picture you supply, which removes the largest source of randomness in video generation.

img2vid · image2video · what is image to video · still to video

Inpainting regenerates the pixels inside a mask you draw and leaves everything outside it untouched. It is how you remove an object, swap a detail, or repair a mistake without re-rendering the whole picture, and the quality of the result depends far more on the mask than on the prompt.

what is inpainting · ai inpainting · video inpainting · generative fill · masked editing

A keyframe is a frame you fix in advance so the model has to pass through it. In AI video the two that matter are the first frame and the last frame, and supplying either one converts an open-ended generation into a constrained one. It is the single biggest lever on consistency.

first frame · last frame · start frame · end frame · first-last frame

Lip sync is the task of driving a face in a photo or video so its mouth matches a supplied audio track. The model repaints the mouth, jaw, and usually the whole lower face frame by frame so the visemes line up with the sound, while the rest of the shot is left as it was.

lipsync · lip syncing · audio driven face · mouth sync

Morphing is a continuous transformation of one image into another, in which features are matched between the two and then interpolated, so the change reads as a single shape becoming another rather than as a cut or a fade. In AI video it is usually produced by generating the frames between two supplied keyframes.

morph transition · morph · tweening · morph cut

Motion transfer copies the movement of a subject in a driving video onto a different character, so a still image or a new identity performs the same action. The pipeline extracts a pose track from the driver, retargets it to the target's proportions, then generates frames conditioned on that track plus your reference identity, which is why it is also called pose transfer.

pose transfer · motion control · motion reference · character animation · driving video

Optical flow is a per-pixel map of how content moved between two consecutive frames, stored as a 2D displacement vector for every pixel. It is the measurement that interpolation, video-to-video, stabilization, and video upscaling pipelines rely on to know what should stay the same across frames, which makes it the machinery behind temporal consistency.

flow field · motion estimation · temporal consistency · flow map

Outpainting extends an image beyond its original borders, generating new pixels outside the frame so they continue the existing scene. It is the same masked fill mechanism as inpainting, just pointed outward: you enlarge the canvas, mark the empty area as the region to fill, and the model paints what would plausibly have been there.

generative fill · image extension · uncrop · canvas expand

Relighting changes the lighting on an existing image or clip without reshooting it, moving the apparent key direction, softness or colour temperature while keeping the subject and framing intact. Doing it convincingly requires an estimate of surface shape, because light has to fall off correctly across geometry the photograph only implies.

ai relighting · portrait relighting · light transfer · relight

Rotoscoping is the practice of isolating a moving subject from its background frame by frame, producing an alpha matte that can be composited over something else. It started as literal tracing over live-action footage and is now mostly done by segmentation and matting models that propose the mask automatically.

alpha matte · roto · ai rotoscoping · ai matting

Style transfer takes the look of one image, its palette, mark-making, texture, and light quality, and applies it to the content of another. Classical neural style transfer optimized a single output against two loss targets. Diffusion pipelines now do the same job from a style reference image or a prompt, which is faster and more flexible but less literal about copying texture.

neural style transfer · style reference · stylization · style matching

Text to music generates a complete audio track from a written description, producing instrumentation, arrangement and often vocals in one pass rather than assembling parts. Models take a style prompt and a duration, and some accept lyrics separately so the words and the arrangement are controlled independently.

t2m · music generation · ai music generation · txt2music

Text to speech converts written text into spoken audio. A modern TTS model does it in two stages, predicting an acoustic representation from the text and then turning that representation into a waveform, which is why the same sentence can be delivered by different voices and at different emotional settings.

tts · t2s · speech synthesis · neural tts

Text to image is the generation task where a model turns a written prompt into a still picture with no reference image. A text to image model samples from what it learned rather than retrieving anything, so the same words with a different seed produce a different but equally valid image.

t2i · txt2img · text2img · what is text to image

Text to video (T2V) is a generation task where a model turns a written prompt into a video clip with no reference image. The model invents the subject, the framing, and the motion at once, which makes it the least controllable and most exploratory of the video tasks.

t2v · txt2vid · text2video · text to video model

The uncanny valley is the dip in comfort that appears when a synthetic face or body becomes almost human but not quite. Viewers accept a stylised character and accept a real person, yet find the near-perfect version between them unsettling. It is the failure mode most AI-generated humans land in.

uncanny valley effect · almost human · near-human

Upscaling increases the pixel dimensions of an image or video. Classic resampling enlarges what is already there, while AI upscaling (super resolution) uses a model to synthesize detail that the source never contained, which is why it can look sharper than the original and also why it can quietly change a face.

super resolution · super resolution ai · ai upscaling · image enlargement · what is upscaling

Vectorization converts a pixel image into shapes described by mathematical paths, so the result scales to any size without softening. An image to svg conversion has to decide where the edges are and how few curves can describe them, which is why it succeeds on logos and flat art and struggles on photographs.

raster to vector · vectorize image · image to vector · svg conversion · autotrace

Video to audio, usually shortened to v2a, generates a soundtrack for a silent clip by watching the picture and producing effects that line up with what happens on screen. It is the automated form of foley, and its hard problem is synchronisation rather than sound quality.

v2a · ai foley · video to sfx · auto foley

Video to video ai takes an existing clip as its main input and re-renders it, restyled, relit, or with objects changed, while the original motion and timing are preserved. The hard part is not the new look, it is holding that look identical across every frame.

vid2vid · v2v · video restyle · video style transfer

Voice cloning builds a synthetic voice from a recording of a real speaker, then reads new text in that voice. Current zero-shot systems need seconds rather than hours of audio: the model extracts a speaker embedding from your reference clip and conditions speech synthesis on it, so nothing has to be retrained for each new voice.

voice clone · speaker cloning · zero-shot tts · voice replication