Anyone experimenting with AI knows that prompting is key to getting the results you need for your project. So let’s go behind the scenes, as we share our own trials and errors, so you can learn from our experts, and maybe even develop new methods.
By the time we reached the final shot of the Veo 3 announcement video, we thought we had it nailed. A cinematic close-up of a Viking warrior, windswept cliffs, a roaring army behind him, the mood fierce and unshakable. We fed the AI what felt like the perfect prompt: vivid lighting, micro-level facial detail, emotional nuance, an entire battlefield’s worth of narrative backstory packed into a paragraph. It should’ve worked.
But it didn’t.
The results came back cluttered, confused, and just felt off. Somehow, all that description left the image feeling overdone and underwhelming. So the creative team decided to change tactics and stripped the prompt down. Way down.
And the result? It worked.
The prompt that didn’t work
Here’s the original, highly detailed prompt that went into our first attempts:
Close-up on the face of a hyper-realistic Viking warrior, lit by a mix of natural daylight and warm, flickering campfire light. Every scar, pore, and skin texture is visible in sharp detail. His weathered face is framed by braided hair and a beard adorned with bone and metal beads. He gazes forward with an intense, penetrating look, the corner of his mouth lifting into a subtle, threatening smile — a balance of calm and power. Behind him, on a rugged coastline with towering cliffs and a roaring sea, stands a massive Viking army, fully armed and still, their unreadable expressions lit by the same firelight. The wind sweeps through their cloaks as waves crash in the distance, gulls wheeling overhead, and the atmosphere is tense, cinematic, and powerful. In a deep, resonant, commanding voice, he says: “So, now we can do whatever we want?”
A rugged coastline with towering cliffs and a crashing sea. A large group of Vikings rushes toward the water, roaring, carrying inflatable pool mattresses, big colorful swim rings in different shapes — donuts, crocodiles, flamingos, and more — along with huge beach balls.
It reads like the setup to a blockbuster trailer. But for the AI, it was just too much — too many lighting cues, too much micro-expression detail, too many atmospheric adjectives layered on top of each other.
The resulting images were disappointing. The Viking looked more like a wax figure than a warrior. And the cinematic power we hoped to highlight? Completely lost. For our team, this was a reminder that detail can sometimes overwhelm AI, even though this can change from scene to scene.
The prompt that worked
Here’s what we used instead — the final, simplified Veo 3 prompt that unlocked the scene we were envisioning:
Close-up on the face of a Viking warrior, lit by natural daylight. He gazes forward with an intense look. Behind him, on a rugged coastline with towering cliffs and a roaring sea, stands a massive Viking army, fully armed and still. The atmosphere is tense, cinematic, and powerful. In a deep, resonant, commanding voice, he says: “So, now we can do whatever we want?”
The army behind him roars and lifts inflatable pool mattresses and giant, colorful swim rings in various shapes — donuts, crocodiles, flamingos, and more.
The Viking warrior smiles.
That small rewrite made a massive difference.
The key? For us, it was giving the AI just enough direction, while leaving space for interpretation. It set the tone and mood, then let the system fill in the visual gaps. In another context, more detail might have been crucial — here, it wasn’t.
Playing with prompt expectations
What this moment taught us wasn’t that simple prompts are necessarily better, but knowing how to be strategic — when to expand and when to contract.
AI video tools like Veo 3 interpret prompts holistically. Over-specifying can sometimes confuse rather than clarify, especially when it comes to tone, lighting, or emotional nuance. On the other hand, too little guidance can leave you with bland, generic outputs.
The takeaway?
Good prompting is not about control. It’s about creative tension.
You’re working with a semi-intelligent system that doesn’t think like you, and that’s the point. AI is not just a tool, but also a creative sparring partner. Sometimes, the best result comes when the AI disagrees with your instinct, and sometimes when it validates it. Both outcomes are valuable. Oftentimes, the best results come when you lean into that push-pull dynamic.
Our advice: Prompt like a filmmaker, not a technician
If you’re exploring generative video, whether with Veo 3, Runway, or any AI platform, here are a few things we learned through trial and error
- Explore layered prompting: In our case, starting detailed and then paring back worked, your path may be the opposite.
- Don’t micromanage: For us, trusting the system to interpret mood delivered better results than over-specifying.
- Be open to surprise: What seems like an error might actually be your best shot.
Final shot, final lesson
That final Viking scene, with its majestic cliffs, inflatable pool toys, and a silent smile breaking into absurd joy, only came to life once we stopped trying to overdirect the AI.
In the end, it was a reminder that just like any creative collaboration, working with AI is unpredictable. What worked for us may not always work, and that’s exactly what makes experimenting with it not just fun, but super valuable.
It’s time for you to put it to the test. Head on over to Artlist’s AI video generator and give Veo 3 a try now.
Did you find this article useful?
