Choosing a Video Model

The video model is the AI that animates your keyframe images into video clips.

Written By Rishikesh from ngram

Last updated About 1 month ago

After the storyboard is created with start and end keyframe images for each scene, the video model generates smooth motion between those frames, producing the actual video footage for your final video.

Where to Find It

The video model selector is in the settings bar below the prompt box. Look for the button labeled with the current video model name.

Video model selector in the settings bar

Click the button to open a dropdown showing all available video generation providers.

Available Providers

Google (Default)

Google offers the Veo family of video models: Veo 3.1 (default), Veo 3.1 Fast, Veo 3.1 Reference, Veo 3 Fast, and Veo 3. These deliver reliable, high-quality video animation that works well across different scene types and visual styles. The "Fast" variants generate quicker at lower cost, while standard models produce higher quality. "Reference" models support multi-frame input for more precise control.

Best for: General-purpose video creation, most use cases, reliable results with reasonable generation times.

OpenAI

OpenAI offers the Sora family: Sora 2 and Sora 2 Pro. These models produce cinematic and polished animations with smooth motion and natural-looking transitions between keyframes.

Best for: Videos that need smooth, cinematic motion; polished product demos; high-production-value content.

Seedance

Seedance offers Seedance 1.5 Pro, Seedance 1 Pro, and Seedance 1 Pro Fast. These provide distinctive motion characteristics and creative animation styles.

Best for: Creative and artistic videos, content where unique motion adds value, experimental projects.

Kling

Kling has a large selection of models: Kling 3 Pro, Kling 3 Standard, Kling O3 Pro/Standard (with Reference variants), Kling 2.6 Pro, Kling O1, Kling 2.5 Turbo, and Kling 2.1 Pro. These offer various approaches to animation with different quality and speed tradeoffs.

Best for: Trying different animation approaches, scenes with complex motion, alternative visual interpretations.

Wan

Wan provides multiple models: Wan 2.6, Wan 2.5, Wan 2.2 Turbo, and Wan 2.1 Pro. These offer different approaches to interpolating between keyframes and can be effective for certain subject matters and movement styles.

Best for: Experimentation, comparing animation styles, specific content types where other models underperform.

How the Video Model Affects Your Final Video

The video model is the last AI step in the visual pipeline. It takes the static keyframe images and produces the moving video clips that make up your final video. Specifically, it determines:

  • Motion quality -- how smooth and natural the animation looks

  • Transition fidelity -- how accurately the video moves from the start keyframe to the end keyframe

  • Visual artifacts -- whether the animation has glitches, distortions, or unnatural elements

  • Generation speed -- how long it takes to produce each scene's video clip

The video model does not affect:

  • The script or narration (that is the text model)

  • The static keyframe images (that is the image model)

  • The voiceover audio (that is the voice setting)

  • The visual style of the images (that is the style setting)

How Video Generation Works

Here is the process that happens behind the scenes:

  1. Script is written by the text model

  2. Keyframe images are generated by the image model (start and end frames for each scene)

  3. Video clips are generated by the video model (animating between keyframes)

  4. Voiceover is generated from the script text

  5. Background music is generated or selected

  6. Assembly combines video clips, voiceover, and music into the final video

The video model handles step 3. All scenes are animated in parallel, which means multiple scenes are processed at the same time to reduce total wait time.

Choosing the Right Video Model

Stick with Google (the default) when starting out. It provides the most consistent results across different types of content.

Consider switching if:

  • You want smoother, more cinematic motion: Try OpenAI for polished animations

  • You want creative or artistic motion: Experiment with Seedance for distinctive animation styles

  • Your current results have visual artifacts: A different video model may handle your specific keyframes better

  • You want to compare options: Generate the same storyboard with different video models to see which produces the best motion for your content

Tips for Better Video Results

  1. Good keyframes lead to good video. The video model works by interpolating between your start and end keyframe images. If the keyframes are high quality and visually consistent, the video will animate more smoothly. Consider adjusting your image model or style if the keyframes are not looking right before changing the video model.

  2. Simpler motion animates better. Scenes with subtle camera movements or gentle transitions tend to produce cleaner video than scenes requiring dramatic motion changes. If a scene has too much going on, consider breaking it into smaller segments.

  3. Check each scene individually. After video generation, review each scene in the storyboard's Video Preview tab. If a specific scene looks off, you can regenerate just that scene without redoing the entire video.

  4. Animation Mode matters too. The Animation Mode setting (found in the Tools menu) interacts with the video model. Make sure you have the right animation mode selected for your content type.

Generation Time

Video generation is typically the longest step in the creation process. Expect:

  • Per scene: 30 seconds to 2 minutes, depending on the model and scene complexity

  • Total time: 2 to 5 minutes for a typical 5-scene video (scenes process in parallel)

  • Progress tracking: The Video Preview tab shows a checklist with checkmarks appearing as each scene completes

Be patient during this step. If you see a message like "Animation interrupted -- scenes still processing in background," the system will automatically resume. Do not navigate away.

Credit Usage

Different video models have different credit costs:

  • Google offers a balanced cost-to-quality ratio

  • Specialized models like OpenAI or Seedance may use more credits per scene

  • The total credit cost scales with the number of scenes in your video

  • Longer duration settings (45s, 60s) typically mean more scenes and more credits

You can track credit consumption at the top of the chat page during your session.

Next Steps