Dont waste your money with bad prompts - Mastering Veo3 prompts
Creators worldwide are stepping up their content game, with tools like Googles Veo3 or ByteDances Seedance Model. Google’s Veo3 platform emerges as a revolutionary solution. It is designed to provide a seamless viewing experience by addressing these technical concerns thoroughly. However, as every other LLM, Veo3 asks for Prompts in order to generate a video.
Given the high costs of roughly 75 cents per generated second and a maximum time of 8 seconds per video, this easily adds up to high Google Cloud bills. 6 EUR for a single video makes you wonder how to improve your prompts in order to generate the right result right away. We have invested hundrets of Euros in generative AI models like ChatGPT, Claude, Gemini and such to gain a lot of experience when working with these tools.
Now before we start - we are proud to be one of the leading agencies in Germany in AI-related questions. We have provided our services to plenty of clients, ranging from our amazingly fast Chatbots to AI consultancy.
So, lets start right into what matters with Google's new Veo3 Model:
How to master Veo3 prompts
A Video are just many pictures
If you want your video to be as good as possible, you need to help the AI to build exactly what you are looking for. What matters most is a strict instruction: Environment, Characters, Set and Setting. Just think of the video prompt as you being an world engineer that stiches together an scene, somehow like God in the early days.
Good thoughts become good videos
Imagine describing your friend a memory of your last vacation. Lets imagine for once a beautiful sandy beach in a Latin American country like Costa Rica. Beautiful sunsets, amazing sand, lots of animals, dusty roads. Now, with this image in mind, try to rephrase it, like you would write it into your journal. Thats also the first important lesson: be strict about your image you want Veo3 to recreate.
Be strict and use constructive data
Nowadays its clear that structured data like JSON is ideal for AIs, as they can greatly work with the structure. That said, what might be a great format for an prompt? Right. An easy one!
{"environment": <...>,"characters": <...>,"storybook": <...>,"style": <...>,}
Now with this structure we can ideally tell Google what we want.
Structure is not everything
Of course, its our thoughts that make the prompt really good. Now, here we have 3 individual keys that we need to fill some information in (or even more, if we want to):
Environment
Lets stay with the beautiful sandy beaches in Costa Rica. How would we describe that to our friend who has never been there. How would you tell your friend how the sand felt on your feet? How the sun burnt your skin? Tell it to Veo3 as well. The more detailled information you give it, the better it will be in replicating that image. Speak about details that should be incorporated. The more details, the better.
Characters / Protagonists
Now, every video usually has a Protagonist, or two. Describe your protagonists, like you have to describe it to someone who is blind. How tall they are. What skin color do they have. Do they speak with accent? How does their hair look like. Eyecolor. Facial features that are standing out. By doing this you can also generate multiple videos with the same protagonists, which helps you even further to create meaningful Veo3 videos for Social Media or your advertising campaign.
Storybook
Probably the most important part, but this time its more of a script. You have already described your protagonists, so now you can refer to them easily. Do they speak with each other? Make it clear in your script. Their conversation can be as easy as a book. This will easily help the AI to understand your story that you would like to tell, but keep in mind that you only have 8 seconds to work with. If you want to have "cuts" in your video, generate more videos and stitch them together with a tool like `ffmpeg` or similar.
Style
The most viral styles are probably the Vlog style. But maybe you want something different - make it clear in your style definition. For example, a very straight forward prompt to generate your Veo3 videos in vlog style is just that: generate the video in the Vlog style". Or explain that the video is done by a cameraman who is not seen (obviously).
Lets get started
Now with this information you are ready to get started with good videos of Veo3. Of course you can still just prompt easily, like "A Dancing cat" and you will get probably just that, Bread and Butter: a dancing cat. However for more advanced generation, this will be very useful for you, especially when using it together with powerful tools like n8n.
Overwhelmed?
New AI technologies might be a bit scary, and every day something new comes up. Thats why its important to stay up to date and work with the newest models when they appear. If you have an upcoming project with AI and need some assistance, feel free to reach out.
