Best Practices
Creating AI video is an iterative process. Unlike a traditional camera, you are “directing” a neural network. Here are the best strategies to get high-quality results with Dal Nulla.
1. Resolution & Duration Strategy
Google Veo has specific strengths depending on the resolution and length of the clip.
| Resolution | Max Duration | Best Use Case |
|---|---|---|
| 720p / 1080p | Up to 8s (extendable) | Standard storytelling, social media, longer narrative shots. |
| 4K (Pro) | Max 8s | High-fidelity hero shots, commercials, detailed close-ups. |
⚠️ Important: 4K generation is computationally intensive. It takes longer to generate and costs more credits. Use it selectively for your most important shots. If you need a video longer than 8 seconds, stick to 720p/1080p to maintain coherence.
2. Achieving Consistency
Consistency is the hardest challenge in AI video. Dal Nulla provides tools to manage this:
- Reference Images: Use the Refs mode in the Scene Inspector. Uploading a character sheet or a specific location image helps the AI “see” what you want.
- Global Context: Use Global Text Nodes to define “immutable” facts about your project (e.g., “The movie takes place on Mars,” “The style is black and white noir”).
- Seed Control: (Advanced) If you like a composition but not the action, try reusing the seed (if available/exposed) or very similar phrasing.
3. Physics and Realism
Veo is excellent at understanding physics, but it helps to be explicit.
- Mention Materials: Describe things as “heavy,” “lightweight,” “viscous,” or “rigid.”
- Fluids & Smoke: Veo excels at fluid dynamics. Prompts like “smoke swirling,” “water splashing,” or “honey dripping” usually yield impressive results.
- Limitations: Complex human interactions (e.g., eating, typing on a keyboard, intricate hand movements) are still prone to artifacts. Try to frame shots to hide hands if they aren’t the focus.
4. Text Rendering
Avoid generating text inside the video. While Veo is better than previous models, AI-generated text is often misspelled or morphs unpredictably.
- Recommendation: Use Dal Nulla to generate the clean video background, then add titles, subtitles, or UI elements in a traditional video editor (like Premiere, DaVinci, or CapCut) afterwards.
- Negative Prompt: Add “text, watermark, subtitles” to your negative prompt.
5. Iterative Workflow
Don’t expect the perfect shot on the first try.
- Draft with Fast Mode: Use the ‘Fast’ model to test prompt ideas and camera movements quickly.
- Refine Prompt: Adjust adjectives and camera commands based on the draft.
- Finalize with Pro: Switch to ‘Pro’ model (and potentially 4K) for the final generation.
- Inpaint/Edit: (Coming soon) Use future tools to fix small glitches rather than regenerating the whole clip.
6. Safety Filters
Google Veo has strict safety filters.
- Avoid prohibited content (violence, hate speech, sexual content, real people/celebrities).
- If your generation fails with a “Safety Filter” error, try softening your language or removing ambiguous terms that might be interpreted as policy violations.