Video
generation.
From a single idea to a finished scene.
With Mr. Grateful
Grateful you're here.
This is AI Mastery.
Last class we made the picture. Today we make the picture move.
Three things.
One home base.
One thinking partner.
Seedance 2.0.
You have the ideas.
Video felt out of reach.
By the end you will generate video three different ways. And you will understand the principles well enough to do it a hundred more times on your own.
- Strange results may show up. That is fine.
- We may get a false restriction on something safe.
- The goal is creative confidence and problem solving.
We build live together. Feel free to make mistakes. That is the whole point.
The opportunity.
And the responsibility that comes with it.
Real founders are
building real businesses
with this.
The power is real. So is the responsibility. Realistic AI humans demand consent, disclosure, and intention behind every video.
The Sacred Act.
Three different ways.
UGC.
Made in Higgsfield Marketing Studio. An AI character talks to camera about the product. Fast, social, and surprisingly convincing.
Creative content.
Image to video, then edited in Premiere Pro. This is where a still becomes motion and a human hand shapes the final feel.
Long-form
commercial.
Multi-edit, built from characters, scenes, and post production. Many shots, many cuts, one story. The deep end of what we'll point at today.
Templates, Seedance, and a few wild prompts.
Different people.
Different intentions.
One.
All the same principles.
It comes down to two things.
Your use case. And your intention.
The framework.
Three ways to generate. Four levels to climb.
Every video starts
one of three ways.
The variables.
- Text
- A call-out shot prompt
- Image
- Character sheet, environment sheet, storyboard
- Video
- Recorded or generated footage to swap and edit
S A S
Subject + Action + Setting. The same prompt that built your images builds your video.
Let's start one generating.
The basic template.
SAS: [Subject] + [Action] + [Setting]
A golden retriever puppy chases a red ball across a sunlit backyard.
You learned this from image generation. Watch how cleanly it carries over.
Add what video gave us.
SACSS: [Subject] + [Action] + [Camera] + [Setting] + [Sound]
A golden retriever puppy chases a red ball, slow dolly tracking alongside it, across a sunlit backyard, with soft grass crunch and a distant lawnmower hum.
Video moves and video speaks. So we add camera motion and sound to the same backbone.
The words that move the camera.
Grow your camera vocabulary.
- Eye Candy, clips organized by cinematic keyword, at eyecannndy.com
- ChatGPT, turn an image you love into a prompt as JSON
- The seedance-director skill, your shortcut
Let it write the
cinematic prompt.
Talk to the skill about your idea. Copy the prompt. Then use your own eyes before you hit generate.
Download the skill fileDirect it beat by beat.
[0–4s]: [Subject] + [Action] + [Setting] + [Camera] + [Sound] [4–6s]: [Subject] + [Action] + [Setting] + [Camera] + [Sound] [6–8s]: [Subject] + [Action] + [Setting] + [Camera] + [Sound]
[0–4s]: Puppy sits still in the grass, static close-up, birds chirping. [4–6s]: Puppy spots the ball and bolts, camera whip-pans to follow, paws thudding. [6–8s]: Puppy catches the ball and shakes it, slow-motion, triumphant little bark.
The skill adds these for you. This is how you control the timing of what happens, and when.
How the pieces fit.
Editing is where the
spirit lives.
We go deep on that in an advanced class. Today, just know it matters.
Cutting, ordering, pacing,
sound design, color.
Everything from here needs images. Characters, environments, products. Let's settle any image-generation questions before we lean on it.
The images that steer a scene.
Combine them
into a storyboard.
Take your character, your environment, and your prop. Feed them in as references and ask the image model to lay them out as a storyboard. Now you can see the whole scene before you ever animate it.
Using the reference images, create a [number]-panel storyboard of [character] in [environment] using [prop]. Show the sequence: [beat 1], [beat 2], [beat 3]. Keep the same character, lighting, and style across every panel.
The Elements tab.
In the upload box, open Elements. Load your images sorted by character, location, and prop. Consistency and easy reuse, in one place.
Add ingredients.
-
01One ingredient.Add a prompt and upload one reference image.
-
02Three ingredients.Story, character, environment, storyboard, call-out prompt.
-
03Swap for control.Trade storyboard and call-out prompt to shape the shots.
- Too many characters at once.
- Environment plus storyboard plus long text.
- Everything crammed into one generation.
Overload the model and it loses the plot. Give it room to stay coherent.
- Skin too glossy? Add "no 3d, no cartoon, no vfx" to force photorealism.
- What you do not specify gets decided for you. Sometimes that is a gift.
- The 180 degree rule. Avoid back-to-back shots flipped a full 180.
Two frames.
The model fills the gap.
Generate a first frame and a last frame. The model invents all the motion in between. This is called interpolation, and it is the highest control you have over a single clip.
Google Omni is the best at this.
Edit real footage with natural language. Swap a character, a place, an object. Kling was the king here, and is still strong, but Omni just took the lead.
One honest catch. The mouth on talking footage is rough. Use it for everything except dialogue. See deepmind.google/models/gemini-omni
Transformations.
Normal, then chaos, then back to normal. A calm scene breaks into something wild, then resolves. The structure carries the surprise.
First-person POV.
A single continuous point of view. Be explicit about what the camera is not doing. No cuts, no zooms, just natural head movement.
Now the easiest thing of all.
Paste a link. Get an ad.
That's it.
That simple.
Same principles.
Your intention.
Your assignment.
Make one video about your company, your product, your service, or your offer. Bring it next time.
The picture moves now.
Go make something.