Skip to main content
An engine is the model that makes your images or clips. You choose it in the “Engine” selector of the tools, or through “Quality” when you create a video. Names change from one season to the next: don’t pick a name, pick a capability. The app shows for each engine what it can do (references, length, sound) and its price.

​
Which capability for which need

​
The capabilities, one by one

Number of reference images. This is the number of images you can attach to guide the result. It goes from zero (the engine works from the text or the start image alone) to several dozen depending on the engine. If you attach too many, the app tells you before sending: “accepts at most N reference images”. Remove an attachment, or pick an engine that accepts more. Strong or weak identity. A strong identity engine keeps a recognizable face from one shot to the next. A weak identity engine drifts: choose it for scenes without a recurring character, settings, objects, or a face that doesn’t need to be exact. For a series, always pick a strong identity. Maximum clip length. It varies by engine: most cap at 15 seconds, some go up to 30. Talking avatar engines go well beyond that, but for a single person to camera. In the tools, the range shown (“X to Y s”) is the reference: stay within it. End image. The clip starts on your start image and ends on a second image. Useful for transformations, transitions, a precise camera move. Not every engine accepts it. Native sound. The engine generates the ambience, even the voice, along with the image. You can turn it off (“Sound off”) to lay your own music afterward. On some engines, sound is paid on top of the image. Lip sync. For a line spoken to camera, the mouth must follow the voice. Some engines invent the voice from the character’s description; others accept an already generated voice and sync the lips to it. Choosing your voice for a character guarantees it across all episodes, in exchange for this sync.
An engine you force in the advanced choice never replaces the lip sync one on speaking shots: the app keeps the engine that knows how to sync lips.

​
Quality: Standard or Premium

When you create a video, “Quality” sets the model that makes the images and clips. Standard takes the base engines, at the best price. Premium takes higher end engines for the chosen style. If the video goes over your cap, the app suggests reducing the length, the number of parts, or switching to Standard.

​
Engines reserved for certain plans

Some engines and some qualities are not included in every plan. In the selector, they carry the name of the plan that unlocks them (“X plan”). If you pick one, the app tells you: “Pick another engine, or change plan in Settings”. The same goes for custom voices, public characters and some autopilot settings.

​
Common mistakes

  • Taking the lowest priced engine for a recurring character: its face drifts. Pick a strong identity.
  • Attaching five references to an engine that accepts none: change engine rather than removing what matters.
  • Asking an engine that doesn’t handle it for an end image: look at the mention before sending.
  • Choosing the most powerful engine for a test: test in Standard, switch to Premium for the final version.
  • Thinking a forced engine changes the lip sync: it keeps its own engine.

​
Settings that go with it

​
Three concrete cases

Take a strong identity, at least one reference (their portrait), an engine capable of lip sync, and native sound or an already chosen voice. This is the case where saving on the engine costs the most: a face that changes from one shot to the next shows right away.
Identity doesn’t matter: a base engine is enough. Look at the maximum length instead if you want a slow shot, and at native sound if you want a generated ambience.
You need an end image. Prepare two images with the same subject (the second one using the first as a reference), then give them as the start image and the end image.

​
How to read the selector

Each line of the “Engine” selector gives a price (per clip, per image or per second, sometimes “from X” when it varies with the resolution), the number of references (“up to N references” or “no reference”), the length range, and the mention “sound” when the engine generates audio. Compare these four lines before comparing the price.