AI Storyboard Generator: turn a story into a run of scene images that match

Paste a story and it breaks it into scenes, then renders one image per scene with the same art style and the same character notes carried through every prompt. Or skip the splitting and hand it your own scene list.

Two things to know first. It runs on your own OpenAI key, so the renders are billed to you by OpenAI on top of the per-image charge here. And the continuity is prompt-based: characters stay recognisable across shots, they are not pixel-identical.

Input	A story in story, or your own list in scenes
Output	One image and one row per scene
Ceiling	20 images per run
Account needed	Your own OpenAI API key
Price	$0.05 per scene image, flat on every plan

🎬 What AI Storyboard Generator does

Two paths in, one path out.

Give it story and a chat model reads it and writes short visual descriptions, one per shot. Give it scenes instead and that step is skipped entirely, which is the cheaper route when you already know your shot list: no chat call, no key needed for splitting.

Either way, every scene description gets the same treatment before it is rendered: your style in front of it, your characters notes after it, and a rule asking for no lettering. That repetition is what keeps the run looking like one piece of work rather than nine unrelated pictures.

Scenes render one at a time, and each one takes a while. A measured two-scene run at 9:16 took roughly 27 seconds per image.

aspectRatio decides the shape: 9:16 gives 1024x1536, 16:9 gives 1536x1024, 1:1 gives 1024x1024.

📥 What you give it

{
  "story": "A lonely lighthouse keeper finds a glowing bottle washed ashore. Inside is a map to an island that should not exist. He sets sail at dawn.",
  "numScenes": 6,
  "style": "cinematic, highly detailed, dramatic lighting",
  "characters": "an old bearded keeper in a yellow raincoat",
  "aspectRatio": "9:16",
  "quality": "high"
}

Nothing in this schema has a default. The form opens with example values filled in, and those are a starting point, not a setting the actor falls back to.

Field	Prefilled with	What it is
story	a lighthouse example	The text to break into scenes. Long input is cut at 6,000 characters before splitting.
scenes	nothing	Your own scene descriptions, one image each. Supplying this skips the splitting step.
numScenes	6	How many scenes the split is asked for, between 1 and 20. Read the note below.
style	a cinematic look	The art direction put in front of every prompt. This is the main lever on how the set looks.
characters	nothing	Who keeps appearing, in a sentence. Added to every prompt so faces and clothes stay close.
aspectRatio	9:16	9:16, 16:9 or 1:1.
quality	high	low, medium or high. This moves your own OpenAI bill, not the charge here.
openaiApiKey	nothing	Your own OpenAI key, used for both the splitting call and the renders. Marked secret.
baseUrl	nothing	Advanced. Any OpenAI-compatible host. Empty means https://api.openai.com/v1.
sceneModel	gpt-4o-mini	The chat model that does the splitting. Ignored when you pass scenes.
imageModel	gpt-image-1	The image model. It has to exist on whatever baseUrl you point at.

numScenes is a request, not a cap. It is what the splitting model is asked for, and the list it returns is used as it comes back, so you can end up with a couple more images than you asked for and pay for those too. The hard ceiling is 20. If you want an exact count, pass scenes yourself.

📤 What you get back

One row per image that was rendered and stored, and the picture itself in the run's key-value store under a key like scene-01-1786411663909.png.

Field	What it is
ok	true. Only delivered images reach the dataset at all.
index	Scene number, counting from 1.
scene	The scene description, either the model's or yours.
prompt	The full text that was actually sent, style and character notes included.
imageKey	The key-value store key the picture is saved under.
imageUrl	A direct link to it.
aspectRatio, size	What you asked for, and the pixel size that came out.
contentType, bytes	The image type as read off the file, and how big it is.

No example row is printed here. Storyboard images are large: a measured two-scene run returned PNGs of 2,257,573 and 1,902,976 bytes.

🧾 Reading the output

The dataset only ever holds finished images. A scene that failed is not in it, and neither is any notice about the run.

Everything else goes to two places: a DIAGNOSTICS record in the run's key-value store, and a separate dataset named storyboard-diagnostics in your account. Both carry a short reason code:

Reason	What happened
missing_api_key	No key was supplied, so nothing was rendered.
no_story_and_no_scenes	Neither field had anything in it.
scene_split_failed	The chat model could not produce a usable scene list.
no_usable_scenes	The split came back, but every scene in it was blank.

The run finishes green in all four cases, with an empty dataset. That is deliberate, since the dataset is what gets billed, but it does mean an empty dataset here means read DIAGNOSTICS, not that your story was too short.

One thing to keep an eye on: storyboard-diagnostics is a named dataset, so it sits in your account after the run and grows with every run you do. Clear it out occasionally.

The Overview table shows index, scene, imageUrl and aspectRatio. prompt, bytes and contentType are in the row but not in that table.

▶️ How to run it

1. Open AI Storyboard Generator and click Try for free.
2. Paste your key into OpenAI API key (BYO).
3. Put your text in Story / script, or your own shots in Scenes (optional).
4. Write a line into Art style, and another into Character notes if people recur. These two do most of the work.
5. Click Start, then take the pictures from the run's Storage tab.

💰 How much does it cost?

$0.05 per scene image. Flat on every Apify plan, no volume tiers.

You pay per picture that was rendered and stored. Scenes that failed are not charged, and a run stopped by one of the four reasons above delivers no images and charges for none. OpenAI bills your own key separately for the splitting call and the renders, and quality moves that figure, not this one.

💡 What people use it for

• Shot sequences for faceless video, where a script has to become pictures before anything is cut.
• Comic and webtoon roughs, working from a scene list rather than a story.
• Ad and pitch concepts, where the point is showing a client the run of shots, not a finished frame.
• Laying out a children's book or an explainer, one page at a time, in one style.

🚧 What it does not do

• It does not lock a face. Continuity comes from repeating your characters notes in every prompt, so people stay close but shift a little shot to shot. If you need a locked character, the reference bank below is the better tool.
• No video. Still images only, no motion, no assembly.
• No editing or upscaling. It renders once. There is no retouch pass and no second attempt on a picture you dislike.
• 20 images per run, whatever numScenes says.
• No text in the images. Prompts ask for none, because rendered lettering usually comes out wrong.
• It will not work without your key, and the key has to be valid for both the chat model and the image model you name.

🧭 Which AI image actor do you need?

If you want	Use
A run of scene images from a story	This one
One character in many poses, locked to a reference set	AI Character Reference Bank
A finished vertical video, script and voice included	AI Faceless Video Generator
Images you already have turned into a video	Storyboard Video Generator
A single thumbnail rather than a sequence	Viral Thumbnail Generator

❓ Questions people ask

Do I need my own key? Yes, for the renders. There is no shared key behind this.

Can I get exactly six images? Pass six entries in scenes. Going through story and numScenes asks for six and usually gets six, but the list is used as returned.

Why did my run finish with nothing in it? One of the four reasons above. Open DIAGNOSTICS in the run's key-value store and it will say which.

Will the same character look the same in every shot? Close, not identical. Write the characters note like a casting brief, with hair, age and clothing in it, and it holds up better.

Can I use a model other than gpt-image-1? Yes, if it answers on your baseUrl. Set imageModel and sceneModel to names that exist there.

How long does a run take? About half a minute per image, one after another.

🆘 If something breaks

Open the Issues tab on the actor page. Include the run ID and the reason code from DIAGNOSTICS, since it already carries the exact message and a suggested fix.