Glossary � AI Video
The language of AI video production.
25 entries. A-Z order. Written to be cited.
A
- AI Lip-Sync
- AI Lip-Sync is the technology that modifies the lip movements of a generated face in real time to perfectly synchronise them with a vocal track or dubbing in any language.
- AI Upscaling
- AI upscaling is the process of increasing a video's resolution (e.g. from 1080p to 4K UHD) through neural networks that actively reconstruct missing micro-details to keep the image sharp.
- AI Voice Cloning
- AI Voice Cloning is the process of creating a synthetic voice identical to that of a real actor, trained on just a few minutes of speech recording.
- Assets Library (SLATE)
- The Assets Library is the SLATE control module that centralises and locks reference images of characters, locations and props to guarantee visual uniformity throughout the video.
B
- Broadcast Audio Mix (−23 LUFS)
- The −23 LUFS audio mix is the loudness adjustment to the European EBU R128 standard, mandatory for broadcasting commercials and programmes on national TV networks.
C
- Camera Control Parameters
- Camera control in AI video models is the ability to simulate real optical movements — pan, tilt, zoom, dolly, crane, focus pull — through mathematical instructions directed at the model.
- Character Consistency
- Character consistency in AI video production is the ability to keep a generated subject's facial features, clothing and proportions constant across different shots and scenes.
- Color Grading in AI Pipeline
- Color grading in the AI pipeline is the chromatic unification and visual stylisation process applied in post-production to give the same photographic "look" to clips generated at different times or with different models.
- Compute (Computational Resources)
- Compute represents the graphics processing unit (GPU) power needed to generate video clips and images through generative artificial intelligence algorithms.
G
- Generative Animatic
- A generative animatic is the first dynamic draft of a video, created by assembling low-resolution AI stills or clips onto the audio track and guide voice, to test the spot's rhythm and direction before final generation.
- Generative Sound Design
- Generative sound design is the creation of sound effects (foley, ambient sounds, audio transitions) via AI models to add audio to a video with no original soundtrack.
I
- Image-to-Video (I2V)
- Image-to-Video is the generative process in which an AI model takes a high-definition static image as its starting point and animates it, transforming it into a dynamic video clip.
- Inpainting and Outpainting
- Inpainting and outpainting are AI editing techniques that allow, respectively, modifying an element within a shot (inpainting) or extending image borders beyond the original boundaries (outpainting).
L
- LoRA Training (Low-Rank Adaptation)
- A LoRA is a mathematical micro-model trained on a specific dataset (e.g. 15-20 photos of a face or product) to teach a generative AI how to reproduce it faithfully in any situation or framing.
M
- Motion Prompts
- Motion prompts are specific text commands inserted into AI model instructions to define the speed, direction of movement and physical behaviour of elements within the shot.
- Multi-Model Workflow
- A multi-model workflow is a production pipeline that selects and combines different generative models (Veo, Kling, Luma, HeyGen) based on the technical needs of each individual shot, rather than relying on a single AI tool.
P
- ProRes 422 HQ Master
- ProRes 422 HQ is a high-bitrate professional video codec developed by Apple, the industry standard for delivering commercials intended for television and cinema.
S
- Script Converter (SLATE)
- The Script Converter is the SLATE module that transforms a traditional screenplay into an automated technical shot-list with metadata ready for generative models.
- SLATE Methodology
- SLATE is MaiDreamsLab's proprietary AI video production system, organised in 5 phases — Script, Layout, Assets, Timing, Export — designed to bring cinematographic set discipline to generative models.
T
- Temporal Consistency
- Temporal consistency is the frame-by-frame stability of visual details (backgrounds, lighting, textures, clothing) throughout the playback of an AI-generated video clip.
- Text-to-Video (T2V)
- Text-to-Video is the direct generation of a video clip from a text description (prompt) alone, which tells the model the desired subject, action, lighting and camera movement.
U
- Uncanny Valley
- The Uncanny Valley is the feeling of unease people experience when facing an AI-generated humanoid figure that is almost identical to a human being, but with subtle unnatural micro-imperfections.
- Unlimited Commercial Licence
- An unlimited commercial licence is the contractual agreement that transfers all usage rights on the AI-generated video to the client, allowing perpetual distribution on every media channel with no impression limits or future royalties.
V
- Video-to-Video (V2V)
- Video-to-Video is a technique where an existing real video is used as a structural or motion reference to guide the generation of a new AI video with a completely different style or subject.
- Visual Drift
- Visual drift is the progressive mutation of stylistic, chromatic or structural details of a scene during the sequential generation of different shots.
Want to apply these concepts to a real project?
Book a free 30-minute diagnostic call with Gianni Spezzano: we'll show you how the SLATE system puts every term in this glossary into practice.