Flawless Glossary
Commonly used terms that you will come across while using DeepEditor.
Blend Shape
Blend shapes are models that represent facial expressions. They are the tools in DeepEditor that enable a user to further adjust the mouth movements of an actor.
DeepEditor
The tool used for vubbing from one language to the same language i.e. just changing the words. For example, replacement of swearing or replacing a line of dialogue with a new line for storytelling purposes.
(DeepEditor's model is based on an English dataset. More languages will be added in the future.)
Draft Vub
A Draft Quality Vub is used for creative decision-making and allows filmmakers to experiment with different storytelling options.
Users can upload offline-quality media, which will be used to generate a Draft Vub. This version can be used during the editing process until you're ready for a full-quality render. You get 3 trial Draft Vubs to start.
To upload a Draft vub, you can use a variety of video and audio specs. See “Asset Specifications” for the full list of media specs. Draft Vubs take about an hour to Generate.
Currently, the output file when you “Download latest render” your new vubbed file will be DNxHR SQ and in the native frame rate and aspect ratio.
Driving Data
The new dialogue to transfer onto your source media.
For more context, visit the Driving Data article.
Euler Angle
The Euler angle (pronounced ‘Oiler’) is a measure of how far the face is turned away from the camera.
0 Euler angle would be face straight onto camera
‘Low Euler’ typically refers to a rotation where we can see the whole face: usually up to 45º, left/right and approximately 15º up/down.
‘High Euler’ therefore refers to a head rotation of more than 45º away from the camera.
Beyond 125º, the visibility of the facial features is limited due to self-occlusion. Commonly the tip of the nose and lips are barely visible by this point
Interpolation
Ranging from 0 to 100, this represents the effect that the Driving Data mouth movement has on the original mouth movement: ‘0' means that there is zero application of the Driving Data (original performance is completely retained), while '100’ would mean that the Driving Data mouth movement entirely replaces the original movements.
Occlusion
An object that blocks another object. One example would be a hand in front of a face. Occlusions can be solid or partly opaque e.g. smoke or rain.
Output Media
The Performance Transfer version of your shot that you can export from DeepEditor.
Performance Transfer (PT) Refinement
The task of refining your generated shot using the Flawless DeepEditor WebApp to refine and perfect the Lipsync of a vub to match its intended Audio. Change framing (e.g. perfect delivery captured in wide angle but desired in close-up)
Phoneme
The smallest distinct unit sound in a language, such as the ‘m' sound in ‘Mary’ or the ‘th’ sound in 'thing’. See Phoneme. From Ancient Greek φώνημα (phṓnēma, “sound”) + -eme.
Plate Proxy
The original source (link to source) video shown within the Refinement tool.
Project
Area to see status, download, or refine your shots.
Refine
The work a user does on a Vub after its been processed by DeepEditor
Visit Refine A Vub for more information.
Scale
Aesthetically this is the ‘performance size’ of the Performance Transfer result. The values can be changed using the refinement tool.
Shot
A continuously recorded series of frames. A shot can also refer to the asset submitted to DeepEditor.
Slapcomp media
Your new generated shot that is viewable in the refinement tool.
Source (When Creating a New Vub)
The original video you submit as the source to generate a new spoken performance of an actor using DeepEditor.
View Source Upload for more context
Source Media (What is displayed in the refinement tool)
The same as the Plate Proxy (link to Plate Proxy) which is the source or plate of the original media shown in the main window.
Standard Vub
A Standard Quality Vub is similar to a Draft Vub but it will Generate a higher quality visual, with less artifacting, but will take longer to do so. Users can upload offline-quality media and it will take about 3-8 hours to Generate.
This version can be used during the editing process until you're ready for a full-quality render.
To upload a Standard Vub, you can use a variety of video and audio specs. See “Asset Specifications” for the full list of media specs.
Currently, the output file when you “Download latest render” your new vubbed file will be DNxHR and in the native frame rate and aspect ratio.
Theatrical Vub (Coming Soon)
A Theatrical Quality Vub is created once filmmakers decide that the Draft Vub meets their needs. At this point, full/online-quality media is provided to DeepEditor, which will then train on this material and render the Vub at high quality.
Full quality media may vary for your needs. Currently DeepEditor can output up to 4k quality videos. See “High-Resolution Output” for more information.
Training data
Training data is very similar to Driving Data. While training data does not direct how your vub will sound, additional information will ultimately produce a better visual result.
See “Create a New Vub” for more information.
Viseme
A viseme is the grouping of any speech sounds that look the same.
Visemes are easiest to think of as the set of mouth shapes that are used in speech e.g. an ‘Oo’ shape or a ‘F' shape, while the ‘B’, ‘M' and ‘P’ phonemes are represented the same 'B/P/M’ viseme, since they have the same mouth shape to generate them.
From Latin vision (“view”) + -eme.
Vub / Visual Dubbing / Vubbing
A Vub is a digitally altered version of a video where specific elements—such as dialogue, facial expressions, or even entire performances—can be modified or replaced while maintaining a high level of realism.
This can be used as both a noun and a verb i.e. you may refer to a ‘vub’ (the result of a visual dub) or the act of ‘vubbing'.