AI video generation
To make an AI video, you don't need a camera, actors or lights. You need an idea, a couple of good reference frames, a video model and the patience for two or three takes. I spent 13 years shooting and running a production company, and in 2023 I built AI into that production. Below is the whole path, step by step, in the order I actually work.
How to make an AI video: 1) define the idea and the goal before you generate anything; 2) write a script with a shot list — a language model can draft it; 3) prepare reference frames (your character, product, location); 4) generate the scenes in a video model such as Seedance 2.5 or Kling; 5) edit the film and add sound and music. Most of the effort goes into steps 1 and 3, not the generation itself.
An AI video is a film where the image, the motion and often the sound are produced by a model from your description and references. You describe the scene in text (the prompt), attach photos or video, and the model generates the shots.
Here's the shift that took me a while to get. When you make video with AI, you become the entire crew: camera operator, writer, gaffer, editor, colorist and prop master. The model draws the pictures. The decisions are still yours.
The most common beginner mistake is to jump straight into generating. Don't. Idea first, then start burning credits.
The goal has to be concrete too. "I want a million views" isn't a goal, it's a side effect. A goal is what the viewer sees and feels, on which platform, in how many seconds.
Where ideas come from:
Once the idea is chosen, you need a shot-by-shot script: what's in each shot, how many seconds, how the camera moves. This is easy to hand to a language model — Claude, ChatGPT, Gemini, any of them will do. Something like: "write an ad script with a shot list for this idea."
Check two things. Can a friend get the point in two sentences? And what makes someone watch to the end? Viewers decide in about a second and a half whether to watch or scroll. A title card at the very start costs you retention immediately.
This is the most labor-intensive step, and the "make a movie in five minutes" videos never mention it. The main problem with video generators is drift: a few seconds in, the character stops looking like themselves. Seedance 2.5 handles this noticeably better, but you still start holding a character with references.
So control starts with images:
How long does this take? On one ad, an hour and a half went into the opening frame alone. After that, everything moves fast.
A tip that saves money: control the generation only where shots and angles repeat. Leave the rest to the model.
Now the video model. Most of the time I work in Seedance 2.5: it's the best at acting with people and it makes a long scene in a single pass. ByteDance released it on July 31, 2026 (TechNode); outside China it's rolling out through Dreamina by CapCut, with clips up to 30 seconds and up to 50 references per generation (CapCut Newsroom), and through the BytePlus ModelArk API (BytePlus). Availability differs by country, so check what's live where you are. The full breakdown is in Seedance 2.5: what it is and how to use it.
But the generator is a detail. If a better model comes out, or you already pay for another service, generate there. The rest of the system stays the same: idea, script, frames, edit.
How to write prompts without wasting credits is a separate article: AI video prompts: how to write them so the model gets it the first time. In short: describe who, where, what they're doing, the camera, the light, the mood and the style; break a complex scene down by seconds. If you're starting from a still image, see how to turn a photo into video with AI.
What to expect from generations:
Video models already generate sound design. If I'm bringing my own music, I write "no music, sound effects only" in the prompt and lay the track in during the edit.
Voice:
The edit is assembling the best takes, color and sound. I cut my films in DaVinci Resolve and hand routine work, like sorting footage, to Claude. Logging 150 clips from my China travel vlog cost me next to nothing.
That's the core workflow, and it's enough to finish your first film. The deeper parts — ad and product films, holding a character across a series — are where I go into detail in my lessons. I post deeper breakdowns and lessons in English on Patreon.
Prices depend on the model and the platform and change every month, so I won't quote a number that's out of date next week. A better rule of thumb: budget two or three takes per scene, not one.
On time: one episode of a micro-drama is one day of work for one person. In my experience that's pocket change compared with a real shoot, where a single episode can cost tens of thousands of dollars.
You can try it for free: many services give daily free credits, and some image generators cost nothing. But free tiers often add a watermark and cap the quality. For regular work you'll need paid access.
For people and long scenes I choose Seedance 2.5. But the generator is easy to swap: if you already pay for another service, the workflow doesn't change.
One generation in Seedance 2.0 at maximum settings took me about four minutes — and with someone else's proven prompt, those four minutes produced a finished ad. An ad with your own character and brand takes longer: most of the time goes into the idea and the opening frames, not the generation.
No, but understanding framing, light and editing helps a lot. Without it, the model will give you a beautiful shot but won't know what the next one should be.
You need a good reference of your face — ideally a video with a head turn. Keep in mind that some models have strict moderation for real faces.
AI video generation · 13 November 2026
How to turn a photo into a video with AI: which photos work, single shot vs. your own storyboard, how to prompt, which model to pick, and the usual mistakes.
AI video generation · 12 November 2026
How to animate a photo with AI: pick the right shot, choose a video model, write a short prompt, keep the face stable and catch the flaws before you post.
Models and tools · 12 November 2026
Seedance 2.5 by ByteDance: what it can do, how it differs from 2.0, where to use it and how to prompt it. A filmmaker's take, based on my own generations.