Apply a Generative Image-to-Video look as a single cinematic frame suggesting emerging movement. Add slight directional blur to loose fabric, hair, foliage, or other movable details, with subtle depth separation and coherent lighting; avoid duplicated anatomy or invented camera angles. Keep the original subject and composition.
Astill image begins to come to life spatially. The camera and figures move. [Additional perspectives emerge].
Principle
Avideo model1[synthesizes asequence of images]from asource image2and, if applicable, text instructions. [Movement arises from learned relationships].
Example
Stable Video Diffusion: 2023. Runway Gen-3: 2024. Gen-4: 2025.
Try it out
Test astarting image with three motion commands. Check camera movement, object movement, and level of detail throughout the entire sequence.
Background
Major public product wave 2023–2025; ongoing development until 2026.
Atrained video model generates atemporal image sequence from astarting image and, if necessary, text instructions. Movement, additional views, changes in lighting, and new details are estimated and synthesized by the model.
Image: Official demo/announcement image. The model completes temporal image sequences from asingle input. The image shows several original demo motifs, not frames from asingle sequence; amotion example can be found on the linked project page.
Distinction
Ken Burns pan or 2.5D parallax move existing image planes. Generative image-to-video can synthesize new views and image content. This is afamily of techniques, not asingle filter. Historical examples are presented without referring to them as the latest models available today.
Aftermath
Astill image could become the starting scene of afictional photograph. The creative brief increasingly shifted towards references, movement guidelines, and selection.
Further examples · 7 images
02Interpolación Generativa AcotadaThis diagram illustrates the process of generative interpolation between a cross and a circle over one second. It demonstrates the fundamental concept of synthesizing intermediate frames to create fluid motion from static endpoints.Olivia Wall · CC BY-SA 4.0 · Wikimedia Commons03AI-generated video of a domestic dragonThe animation shows a woman interacting with a dragon. It demonstrates the model's ability to maintain subject consistency while generating complex organic movement and camera panning from a high-detail initial frame.prompted by Lumi's AI Dreams · Public domain · Wikimedia Commons04AI scare ad Fidesz warA low-resolution video still shows a girl looking out a window. It demonstrates the characteristic texture and slight spatial warping of early generative AI video models used in 2024 political advertising.Fidesz Budapest · Public domain · Wikimedia Commons05Coca-Cola AI ad - truck with misspelled logoThe image shows a truck with a garbled brand logo and flickering lights. This displays the common text-synthesis errors and temporal inconsistencies found in generative video outputs from late 2024.Silverside AI for The Coca-Cola Company
AI technologies used in the ad campaign include: Leonardo, Luma, Runway, Kling, · Public domain · Wikimedia Commons06AI illustration: A split frame compares a static oil painting to a modernized, fluid versiA split frame compares a static oil painting to a modernized, fluid version of the same scene. This highlights the synthesis of new, dynamic pixels from a fixed source image.Illustration generated for the Atlas (AI)07AI illustration: A side-by-side comparison shows a static city scene gaining motion blur aA side-by-side comparison shows a static city scene gaining motion blur and dynamic lighting. This illustrates how generative models synthesize movement and temporal changes from a single base frame.Illustration generated for the Atlas (AI)
