Video Generation

This guide covers how to generate AI videos from keyframe images, compare multiple Takes, and select the best result.

Overview

Cutflow's video generation is an Image-to-Video approach where the AI creates a moving video using the keyframe (representative image) as the first frame. Since the selected keyframe becomes the starting frame of the video, you will get the best results by finalizing the desired composition and character at the keyframe stage before generating video.

Video generation is processed asynchronously and takes approximately 2-6 minutes depending on the model. You can continue working on other cuts while generation is in progress.

Generating Video from a Keyframe

Prerequisites

To generate a video, a representative keyframe must be selected for that cut. The video generation button is disabled when no keyframe is selected.

Generation Steps

Verify keyframe selection: Confirm that a representative keyframe is selected in the cut editor.
Click the Generate Video button: Click the "Generate Video" button in the Take area at the bottom of the cut.
Configure options: Select the video duration, audio generation option, and AI model.
Confirm credits: The estimated credits based on the selected model and duration are displayed. If the balance is insufficient, generation is blocked and a recharge notification is shown.
Start generation: After confirmation, generation begins. Progress is displayed on screen.

Asynchronous Processing

Video generation is processed asynchronously on the server.

Once generation starts, the status is displayed as processing.
A progress indicator appears along with the estimated completion time.
When generation is complete, the video loads automatically.
You can edit other cuts or perform other tasks while generation is in progress.
Estimated processing time by model: Fast models ~2-3 minutes, Standard models ~2-4 minutes, Premium models ~5-6 minutes

Duration & Audio Options

Supported Durations

Video duration support varies by model. Not all models support all durations.

Model	Supported Durations
Seedance 1.5 Pro	4, 6, 8, 10, 12 seconds
LTX 2 Fast	6-20 seconds (in 2-second increments)
Wan 2.6	5, 10, 15 seconds
MiniMax 2.3 Fast	6, 10 seconds
MiniMax 02 Pro (default)	6, 10 seconds
Kling 2.6 Pro	5, 10 seconds
Veo 3.1 Fast	8 seconds (fixed)
Kling O3	3, 5, 10, 15 seconds
Veo 3.1	8 seconds (fixed)

If the selected model does not support the chosen duration, only compatible models will be available for selection.

Audio Generation

Some models can automatically generate audio (ambient sounds, sound effects, etc.) along with the video.

Model	Audio Support
Seedance 1.5 Pro	Supported
LTX 2 Fast	Supported
Kling 2.6 Pro	Supported
Veo 3.1 Fast	Not supported
Kling O3	Not supported
Veo 3.1	Not supported
Wan 2.6	Not supported
MiniMax 2.3 Fast	Not supported
MiniMax 02 Pro	Not supported

For scenes that require audio, select an audio-supported model. Enabling the audio filter during model selection will show only audio-supported models.

Duration-Based Credit Multiplier

Video credits are calculated as base credits x Duration Multiplier. Credits increase proportionally as the duration gets longer.

Example (Seedance 1.5 Pro, base 12 credits):

Duration	Multiplier	Credits Consumed
4 sec	0.8x	~10 credits
6 sec	1.0x	12 credits
8 sec	1.3x	~16 credits
10 sec	1.6x	~19 credits
12 sec	2.0x	24 credits

For detailed model-by-model credit information, see the AI Models and Credits documents.

Take Management

What is a Take?

A Take is each version of a video generated from the same keyframe. Even with the same keyframe and prompt, AI generation produces different results each time, so you can generate multiple Takes to compare.

Unlimited Regeneration

You can generate as many Takes as you want. Each Take is charged individually, and previous Takes are preserved (not deleted).

Comparing Takes

All generated Takes are displayed in a grid in the Take area at the bottom of the cut.
Click each Take to play it in the preview screen.
Information such as the model used and generation time is displayed for each Take.

Selecting the Representative Take

Select the most appealing video as the representative Take from multiple Takes.

Click the check (confirm) button on the desired Take.
The selected Take is designated as the final video for that cut.
The representative Take is used during sequence rendering (export).
You can switch to a different Take at any time.

Checking Failure Reasons

When video generation fails:

The Take card displays the failure status and reason.
Common failure reasons: model server overload, content safety filter triggered, network timeout, etc.
Credits are not charged for failed generations (video generation uses a post-success billing model).
You can use the Retry Check button to re-query the generation status.

Video Prompts

AI Auto-Generated Motion Prompts

Cutflow automatically generates a video prompt (Video Prompt) using AI based on the cut's script.

It analyzes the script content to generate a prompt focused on dynamic elements such as character movements, facial expression changes, and camera movements.
@charactername references in the prompt are automatically replaced with the character's appearance description.
It uses an LLM (Gemini 2.5 Flash) for generation, and prompt generation itself is free.

Custom Prompts

You can edit the auto-generated prompt or write a video prompt from scratch.

Elements of an effective video prompt:

Character movement: "@Jaden slowly lifts his head and looks out the window"
Expression change: "Transitioning from a surprised expression to a gradual smile"
Camera movement: "Dolly in, slowly approaching the face"
Scene transition: "The sound of rain grows louder as the screen gradually darkens"
Time passage: "Morning sunlight gradually fills the room"

Camera Movement Keywords

Key camera movements you can use in video prompts:

Keyword	Description
pan left/right	Camera moves horizontally left or right
tilt up/down	Camera moves vertically up or down
dolly in/out	Camera approaches or retreats from the subject
zoom in/out	Zoom in/out
tracking shot	Camera follows the subject
crane shot	Camera moves from high to low or low to high
static	Fixed shot (no movement)
slow motion	Slow motion effect
handheld	Handheld camera feel (slight shake)

You can ask the AI chat assistant "Create a video prompt for this cut" to automatically generate a video prompt tailored to the script context and apply it immediately. For more details, see the AI Chat document.

Tips

Start with shorter durations: Generate at 4-6 seconds first to check the style and motion, then generate at a longer duration if satisfied. This saves credits.
Keyframe quality matters: Since the keyframe is the first frame of the video, a high-quality keyframe leads to better video results.
Write prompts focused on motion: Video prompts that focus on movement, changes, and camera paths rather than static scene descriptions produce more natural results.
Try a different model if generation keeps failing: If a model fails repeatedly, switch to a different model and try again.

FAQ

Q: How long does video generation take?

It depends on the model. Fast models take approximately 2-3 minutes, Standard models approximately 2-4 minutes, and Premium models approximately 5-6 minutes. You can continue working on other tasks while generation is in progress.

Q: Are credits charged if video generation fails?

No. Video generation charges credits after success. No credits are charged for failed or timed-out generations.

Q: Can I generate a video without a keyframe?

In the current version, selecting a keyframe (representative image) is required. Text-only video generation without a keyframe is planned for a future update.

Q: Can I delete previously generated Takes?

Yes, you can click the delete (X) button on each Take to remove unwanted Takes. However, a Take selected as the representative must be deselected before it can be deleted.