Urgent.News

What's breaking now, across hundreds of outlets.

AI

How to write prompts for AI video models: a six-part formula (Seedance, Kling, Veo)

Video models like Seedance 2.5, Kling 3.0 and Veo 3.1 understand plain text, but a short prompt such as "a girl dances on the street" almost always gives a random result: the camera jitters, the motion is wrong, and the lighting changes on every run. Below is a six-part formula that makes generation predictable, a way to test prompts without burning credits, and three before/after examples. The…

Writing effective prompts for AI video models like Seedance 2.5, Kling 3.0, and Veo 3.1 is crucial to achieving predictable results. A short prompt like "a girl dances on the street" often leads to poor outcomes, such as camera jitter, incorrect motion, and inconsistent lighting. This guide presents a six-part formula to create stable and reliable video prompts, along with three before/after examples and tips for testing prompts without burning credits.

Key differences exist between image and video prompts. In an image, the model decides what is in the frame; in a video, it must determine what changes over time, the camera move, and any accompanying sound. Failing to provide this information results in unpredictable outcomes due to the model's "fill-in-the-blanks" process. A good video prompt should describe not only the image but also the action over time.

The six-part formula for crafting video prompts includes:

1. Subject: A single main subject and 2-3 details. For example, "a young woman in an oversized white hoodie."

2. Action: One action described for the entire clip. For instance, "dances to an upbeat beat."

3. Place: The location and time of day, such as "neon-lit Tokyo street at night."

4. Camera: One camera move per clip, like a "slow orbit around her."

5. Light and style: Lighting, texture, and genre details, like "rain reflections, cinematic light."

6. Sound: If audio generation is enabled, describe the beat and city sounds.

A few essential tips include keeping the subject and action concise, including the setting to establish background and lighting, and placing style and lighting details at the end of the prompt. Camera vocabulary plays a vital role, with options including static shot, slow dolly in/out, orbit around the subject, tracking shot, handheld, low/high angle shots, and more.

To test prompts without incurring costs, draft them first, using the lowest resolution and shortest duration when possible. Fix the seed when available to isolate the prompt's impact. Change one element at a time and keep a log of your results. Remember that costs can vary significantly between plans, so selecting an appropriate plan is essential.

Finally, avoid common pitfalls like including text in the frame, close-ups of hands with complex movements, multiple actions in a short clip, conflicting styles, or excessive adjectives without an action. By following this formula and checklist, you can significantly improve the predictability and quality of your AI-generated video content.

Rewritten by urgent.news; may contain errors — check the original.

Read the original at dev.to →

Report a problem with this story

More in AI

I built an open-source AI gateway for Kubernetes: budgets, kill switch and evals in CI

Every team wants AI. Then three questions arrive: What does it cost? Where are the keys? Can we stop it? I'm a platform engineer, and I built SanVi to answer all three on Kubernetes.

  • Sandhilk created open-source AI gateway called SanVi for Kubernetes
  • SanVi manages budgets, kill switch, and evaluations in CI
  • System logs all calls with team, model, tokens, and costs in Postgres

How to Monitor Agents Built with Pydantic AI or Google ADK

To monitor agents built with Pydantic AI or Google ADK, export the OpenTelemetry traces both frameworks already emit to a monitoring backend, and attach a user and session ID to each run.

  • Export OpenTelemetry traces to monitoring backend
  • Use Agnost AI for conversation monitoring and alerts
  • Pydantic Logfire inspects Pydantic AI runs

More from Saturday 21 November →