Fix both ends, and generate the middle

Start-End Frame Video is the third entry in the generation type list.

Start-End Frame Video sits between Text to Video and Draw to Video in the type dropdown. Selecting it rebuilds the form with two required image blocks instead of one.

That second required input is the whole distinction. Image to Video treats a last frame as optional, and here both ends of the shot are inputs the generation cannot run without.

The roster is short. Vidu Q3 Pro is the default at 540p to 1080p and 1 to 16 seconds, and Seedance 2 is the alternative at 720p to 1080p and 4 to 15 seconds.

Vidu Q3 Pro is the more flexible of the two on length, reaching both shorter and longer than Seedance 2. Where a transition needs to be very brief, a one second Vidu generation is available and Seedance will not go below four.

Two models here, against thirteen in text to video.

The start frame accepts files up to 300 MB.

The start frame block takes an upload, a previous completed task, a saved AI Actor, a saved AI Character or a direct image URL. It accepts JPEG, PNG, WEBP, GIF and AVIF at a minimum of 300 by 300 pixels.

Its upload ceiling is 300 MB, which is three times what the end frame block allows. If you are working from large source files, the asymmetry is worth knowing before you hit it.

The end frame block mirrors the start frame block, with the same five input methods and the same accepted formats. Its upload limit is 100 MB rather than 300 MB.

The two images need to be plausibly the same scene for the transition to read as a shot rather than a dissolve. Frames that share a subject, a lighting setup and an approximate camera position give the model a path to interpolate along.

The end frame takes the same five inputs, with a 100 MB ceiling.

The prompt guides the route between the two frames rather than the frames themselves.

Both ends are already fixed by your images, so a prompt describing what they contain is spent on information the model already has. The useful content is what happens in between.

Naming the route matters when more than one is plausible. Two frames of the same room from different angles could be a pan, a cut on movement or a dolly, and saying which one you want is the difference between three attempts and one.

The slider runs from 1 to 16 seconds on Vidu Q3 Pro and starts at 5. Duration has an unusually direct effect here, because it sets the speed of the transition rather than just the length of the clip.

The same two frames over 2 seconds and over 12 seconds are different shots. A short duration reads as a snap or a whip, and a long one as a slow reveal, so this control is a creative decision rather than a technical one.

Duration from 1 to 16 seconds on Vidu Q3 Pro, defaulting to 5.

Movement amplitude, a control that exists here and not in image to video.

Movement Amplitude offers Auto, Small, Medium and Large. It governs how much motion the model introduces on the way between your two frames, separately from how far apart those frames are.

Auto is a reasonable default. Small is the setting to reach for when the two frames are already close and the automatic choice is adding movement the shot does not need, and Large suits a transition that should feel energetic rather than smooth.

Vidu Q3 Pro offers 540p, 720p and 1080p, defaulting to 720p. Seedance 2 starts at 720p and does not offer the 540p option.

540p exists to make testing cheap. Whether two frames interpolate into a coherent shot is visible at 540p, so establishing that first and regenerating at 1080p is the sensible order.

Three resolutions on Vidu Q3 Pro, with 720p selected by default.

Two independent audio toggles, both on by default on this model.

Two separate toggles control audio. Generate Audio produces sound tied to what is happening in the picture, and Background Music adds a score underneath it.

For a clip heading into an edit with its own sound design, turning both off avoids generating audio you will discard. For a standalone piece the generated track saves a step.