September 5, 2026
Making your storyboard in AI before paying for the video: the method that divides the bill by ten
An image costs 1 to 5 credits, a video shot costs 26 to 96. How to validate framing, character and light in stills, then animate only what is already approved, with the full costing of a thirty second film.
Frank HoubreFounder of Imaginode
The price ratio that should decide your method
An image costs between 1 and 5 credits while a five second video shot costs between 26 and 96: correcting in image is ten to twenty times cheaper than in video.
Put those two numbers side by side and the whole method follows. An image on Flux Schnell costs 1 credit, on Flux 2 Klein 3 credits. A five second video shot starts at 26 credits on Kling 2.5 Turbo and climbs to 48 on Seedance 2.0 Fast, or 96 for ten seconds.
In other words, a framing mistake discovered in video costs you thirty times what it would have cost in image. And believe me, framing mistakes are never discovered before you have seen the image.
That is the whole point of the storyboard, a tool cinema has always used for a very down to earth reason: it costs less to draw a shot than to film it. AI invented nothing here, it just made the thing available to someone who cannot draw.
BalanceWhat a storyboard really has to contain
A useful storyboard fixes, for each shot, the framing, the subject, the action and the duration: those four pieces of information are what will let you generate the video without hesitating.
A storyboard is not a pretty board of drawings. It is a decision document. For each shot it answers four questions: what do we see, how is it framed, what happens during the shot, and how long it lasts.
Those four answers are exactly what you will have to give the video model. A well made storyboard is your video prompt already written, plus the starting image the model will animate. Nothing is lost between the two stages.
And it makes visible what no text shows: rhythm. Six thumbnails laid side by side tell you immediately whether your film breathes or whether it strings the same shot together three times. That is the kind of thing you never see in a table of text.
Start with the shot list, in a Text node
First write one sentence per shot, in order, with a duration: the written shot list takes ten minutes and stops you generating shots that serve no purpose.
Before a single image, open a Text node and write your film in sentences. One sentence per shot, in order, with the target duration. Shot one, three seconds, the hand putting the cup down on the counter. Shot two, four seconds, the customer's face looking up. You get the idea.
The Text node can write it for you too: give it an instruction like write the shot list for a thirty second film about this product, three sentences, one per shot, no title and no numbering. A language model costs 1 to 3 credits per generation, which is negligible at this stage.
But reread it and decide yourself. A model will happily propose eight shots where six would do, because nothing pushes it to cut. The shot list is the only moment when you can delete a shot for free: from the thumbnail onwards, every deletion is budget already spent.
Generating the thumbnails: the right model and the right budget
Explore at 1 credit on Flux Schnell to find the composition, then lock the chosen thumbnail at 3 credits on Flux 2 Klein: six validated shots cost under fifty credits.
Exploration phase: Flux Schnell, 1 credit, without counting. You are looking for the composition, not for beauty. Three or four tries per shot is generally enough to see what works. Twenty four tries for six shots makes 24 credits, about twenty cents.
Locking phase: the chosen thumbnail is regenerated cleanly on Flux 2 Klein at 3 credits, or on Seedream 5 Lite at 5 credits if you need it to accept reference images. Six final thumbnails, between 18 and 30 credits.
Total for a complete six shot storyboard: around fifty credits, roughly fifty cents. Compare that with the 288 credits six video shots generated blind on Seedance 2.0 Fast would cost, half of which you would throw away.
Holding the character across the six thumbnails
A Reference node called by @mention in each prompt keeps the same face on every thumbnail, where repeating the description in each prompt fails.
As soon as a human appears in more than one shot, the storyboard reveals the very problem it was meant to anticipate: it is not the same person from one thumbnail to the next. Better to settle it here, at image prices, than in video.
Create a Reference node, give it a name, a description and three photos of your character, then call it with an @mention in all six prompts. The images go off with the prompt and the model copies real proportions instead of inventing new ones each time.
Check that the model you picked accepts references: the port does not appear on the node when it does not, and that is the only indicator. The complete method, including how to choose the starting photos, is in consistent character in AI and in creating an AI character.
Holding the light and the art direction
A Style node plugged into the six nodes imposes the same palette and the same treatment everywhere, and is changed in a single place.
Second thing the storyboard reveals early: six shots that do not share the same light do not make a film. One is in warm end of day light, the other in cold neon, and the edit will look like a patchwork.
The Style node plays that part. You describe the art direction once, you add inspiration images to it, you plug it into the six Image nodes. All your thumbnails inherit from it, and the day you change your mind about the palette, you change it in one place.
It is also what makes the storyboard presentable. Six coherent thumbnails look like a film; six disparate ones look like tests. When the document goes off to a client, the difference is enormous.
The Camera node: putting the staging into the storyboard
Setting framing, lens, aperture and angle in a Camera node fixes the staging from the thumbnail on, and the same settings then travel to the video.
The storyboard exists to decide the framing, and the Camera node is made exactly for that. You choose the shot type, the lens, the aperture, the angle, and it translates those settings into the language the model understands.
The benefit is twofold. First you get thumbnails that really look like what the shot will be, with the right compression of an 85 mm or the right background blur at f/1.4. Then that node plugs straight into the Video node: the staging decided in image goes to video without being retyped.
A single Camera node can feed several shots to keep one lens throughout, or you can place one per shot if your shot list alternates wide and close up. The detail of the settings and what they change is in directing the camera in AI.
The moment the storyboard becomes the video
Each validated thumbnail becomes the starting image of a Video node: the model animates an already approved composition instead of inventing one.
This is the step that justifies all the rest. A validated thumbnail plugs into a Video node and becomes its first frame. The model no longer composes, it animates. You have removed the most uncertain variable from the equation.
In practice that means the video prompt becomes very short, and that is a good thing. The image already says what the scene looks like. All that is left is to describe the movement: the camera moves slowly forward, the hand closes, the eyes look up. Three words per shot.
The rate of usable shots jumps, because most video failures come from framing or from the character, not from movement. And when a shot still fails, you rerun that one shot keeping the same starting image, for 26 to 48 credits.
The full costing of a thirty second film
A fifty credit storyboard then six video shots at 26 credits: a thirty second film comes to around 200 credits, editing included since editing is free.
Let us total it up on a concrete case, six shots of five seconds. Exploration in images, around twenty tries at 1 credit: 24 credits. Final thumbnails, six at 3 credits: 18 credits. The complete storyboard is 42 credits.
Video generation, six five second shots in 720p on Kling 2.5 Turbo at 26 credits: 156 credits. The Editing node assembles it all and the export is computed on your device, without spending a credit. Total: around 198 credits, say two euros.
Compare that with the method without a storyboard. You generate straight to video, you throw half of it away because the framing is wrong, and you redo six. Twelve shots at 26 credits make 312 credits, and you have spent twice as long. The complete method for going beyond ten seconds is detailed in AI video longer than 10 seconds.
The storyboard as a client deliverable
Getting six thumbnails approved before producing the video moves the corrections to the cheapest side and secures the order.
There is a commercial reason for all this, and it is worth at least as much as the credit savings. A client who discovers the finished film has remarks. That is normal, it is their project. Except that at that stage, every remark costs a video regeneration.
Send them the six thumbnails before producing. They will tell you the third shot is not right, that the light is too cold, that they would prefer a wider shot on number two. Those corrections cost 3 credits each and ten minutes.
And the agreement obtained on a storyboard protects you afterwards: what was approved was approved. On how to price these different stages in a quote, I detailed my rate card in invoicing AI creation to a client.
What the storyboard does not settle
A perfect thumbnail guarantees neither the movement, nor that the character holds during the animation, nor the continuity between two shots.
I would be dishonest to let you believe the storyboard solves everything. It fixes the starting point of each shot, not what happens during the five seconds that follow.
A video model can perfectly well start from a flawless image and let the face drift by the third second, or invent a camera move you did not ask for. That is the remaining share of chance, and it still exists.
What it does settle is all the rest: framing, composition, light, style, the character's identity, the rhythm of the shot list. In other words the overwhelming majority of the reasons a shot gets thrown away. That is already enormous, and it costs 42 credits.
The procedure, in five steps
Written shot list, exploration at 1 credit, locked thumbnails, animation from the thumbnails, free editing: five steps in that order, never another.
One, the shot list written in a Text node, one sentence per shot with its duration. Two, exploration in images at 1 credit, without holding back. Three, locking the thumbnails on a good model, with a Reference node for the character, a Style node for the art direction and a Camera node for the framing.
Four, the animation: each thumbnail becomes the starting image of a Video node, with a short prompt that only describes the movement. Five, the edit, which assembles, trims, adds transitions and sound, for free.
If you have never generated video, start with your first AI video to get to grips with a single Video node, then come back here for the complete method. And for a product advert, the step by step path is in product video ad.