09 — AI Video Generation
AI Video Generation is available under AI Generate → Video. Generated clips can be downloaded or added directly to your timeline as video segments.
A Replicate API key is required. Some models (Sora, Grok) may additionally require their own provider API keys.
Available Models
| Model | Provider | Notes |
|---|---|---|
| Runway Gen 4.5 | Runway ML | Top-rated quality; strong prompt adherence |
| Sora 2 | OpenAI | High quality; supports image conditioning |
| Sora 2 Pro | OpenAI | Enhanced version of Sora 2 |
| Veo 3.1 | Excellent motion quality | |
| Veo 3.1 Fast | Faster version; slightly reduced quality | |
| Grok Imagine Video | xAI | Creative and stylistic outputs |
| Kling V3 | Kuaishou | Strong realism |
| Kling V3 Omni | Kuaishou | Enhanced Kling with broader motion range |
| Hailuo 2.3 | MiniMax | Fast generation; good for B-roll |
Controls
| Control | Description |
|---|---|
| Prompt | Describe the video scene, motion, and style |
| Aspect Ratio | 16:9 (landscape), 9:16 (portrait/short-form), 1:1 (square), or custom dimensions |
| Duration | Clip length in seconds; range varies by model (typically 5–15 seconds) |
| Guidance Scale | How literally the model follows your prompt |
| Steps | Sampling steps; higher = more quality, slower generation |
| Seed | Set for reproducible outputs; leave blank for random |
Optional: Reference Image
Condition the video on a static image. The model uses the image as a visual starting point or style reference.
- Click Add Reference Image.
- Enter a publicly accessible image URL.
Optional: First Frame
Set an exact first frame for the video. Some models support specifying a starting image that the video begins from and animates away.
Generating a Video
- Select your model.
- Write a descriptive prompt (include motion verbs — e.g., “slowly pans left”, “zooms in on”, “water flows gently”).
- Set aspect ratio and duration.
- Optionally add a reference image.
- Click Generate.
- The job is added to the Job Queue. Video generation typically takes 1–5 minutes depending on the model and duration.
Tips for Better Results
- Be specific about motion. Models respond well to explicit direction: “camera slowly tilts upward” rather than just “mountain landscape.”
- Specify lighting and atmosphere. “Golden hour sunlight, warm tones” gives the model more to work with than a basic scene description.
- Keep prompts focused. One clear scene per generation produces better results than a complex multi-scene prompt.
- Use short clips and sequence them. Generate multiple 5–10 second clips and arrange them on the timeline rather than trying to generate one long clip.
Using a Generated Video
From the Results tab or Recent Results:
– Download — saves the MP4 file to your computer
– Add to Timeline — inserts the clip as a video segment at the current playhead position
Previous: AI Image Generation | Next: Talking Head / Lip-Sync →