Storyboard AI: From Shot List to Consistent Frames and Video

Storyboard AI turns a shot list into frames, and the frames into an animatic. We ran the whole pipeline on one 4-shot story: a character sheet, the same shot list through three image models, then the frames through five video models. Our pick: Seedream for the board, the only model that kept every prop in all four frames, and Kling for the clips. That took the story from a blank page to a 20-second animatic for $1.51.

Below: the shot list template we used (copy it), which image model kept the character and the props, what five video models did with the same frame, what each step cost, and how to run it all from one API key.

Step	Best in our test	Time per call	Billed
Character sheet	Nano Banana 2.1	17 s	$0.03
4-frame board	Seedream 5.0 Lite Edit (one call)	252 s	$0.14
Board on one page	Nano Banana 2.1 Edit	25 s	$0.05
4 frames to 5 s clips	Kling V3 I2V	67 to 72 s	$1.34 for 4
4 frames, one 15 s take	Vidu Q2 Turbo Multi-Frame	136 s	$0.018 to $0.08 per s (list)

How Storyboard AI Works: Shot List, Frames, Animatic

A storyboard is a shot list you can see. Dedicated apps such as Boords, StoryTribe, and Storyboarder.ai wrap the steps in one interface; underneath, every AI storyboard does the same three things:

1. Shot list. Each shot gets a size (wide, medium, close-up), an angle, the action, and a duration. This is the part you write.
2. Frames. An image model draws each shot from a character reference, so the hero looks the same in every panel.
3. Animatic. An image-to-video model turns each frame into a short clip, and the clips cut together into a rough cut you can time and pitch.

The step people skip is the first one, and it is the one that decides whether frames 2 to 4 still match frame 1.

The shot list template we used

Our story, "The Lantern Run": a small delivery robot carries a paper lantern through a rainy night market to an old tortoise's shop. Four shots, five seconds each.

"What we see" goes into the image model; "Motion and camera" goes into the video model later. Keeping them apart is what lets one shot list drive both steps. The same list as JSON, ready for a script or an agent:

{
  "character": "Bolt, a small round mustard-yellow delivery robot about knee high, one large round blue camera eye, a red knitted scarf, a short antenna with a green tip, two thin arms with three-finger grippers, and a single rubber wheel",
  "style": "2D animated feature film still, soft painterly backgrounds, warm orange lantern light against cool blue rain, gentle film grain.",
  "shots": [
    {
      "n": 1,
      "size": "Wide establishing shot",
      "angle": "high angle",
      "action": "Bolt rolls into a crowded rainy night market alley carrying a glowing paper lantern, strings of lanterns overhead, puddles reflecting the light.",
      "motion": "The small yellow robot rolls forward through the puddles, rain falls, lanterns sway.",
      "camera": "Slow push in from high angle.",
      "duration": 5
    },
    {
      "n": 2,
      "size": "Medium tracking shot",
      "angle": "side view at robot height",
      "action": "Bolt weaves between food stalls, steam rising from a noodle stall, the lantern held high to keep it dry.",
      "motion": "The robot weaves left and right between stalls while steam drifts past.",
      "camera": "Camera tracks sideways, keeping pace with the robot.",
      "duration": 5
    },
    {
      "n": 3,
      "size": "Close-up insert",
      "angle": "eye level",
      "action": "Bolt's three-finger grippers cradle the paper lantern, raindrops beading on the paper, the blue eye reflected in it.",
      "motion": "Raindrops roll down the lantern paper, the grippers tighten slightly, the lantern flame flickers.",
      "camera": "Static camera with a slow rack focus from the raindrops to the eye reflection.",
      "duration": 5
    },
    {
      "n": 4,
      "size": "Medium wide shot",
      "angle": "low angle",
      "action": "Bolt stops at a small shop doorway and lifts the lantern to an old tortoise shopkeeper in a cardigan, warm light spilling out.",
      "motion": "The robot lifts the lantern up, the tortoise smiles and reaches for it.",
      "camera": "Slow tilt up from low angle.",
      "duration": 5
    }
  ]
}

Turn a script into a shot list

Any capable LLM can do the breakdown. This prompt returns the format above; the line that matters most is the last one, because prop drift was the most common failure in our test.

You are a storyboard artist. Read the script below and break it into a shot list of 4 to 8 shots. Return JSON with: character (one sentence that fixes every visual trait: body, colors, clothing, props), style (one sentence), and shots, where each shot has n, size (wide, medium, close-up, insert), angle (high, low, eye level, side), action (what is in the frame, one sentence), motion (what moves in the clip), camera (one camera move), and duration in seconds (5 or 6). Name the same prop the same way in every shot.

Script: <paste your script here>

Storyboard AI Test: Same Shot List, 3 Image Models

Lock the character first

Before any panel, we generated one character turnaround sheet with Nano Banana 2.1 (17 seconds, $0.03) and passed it as the reference image to every frame. One sheet with front, side, and back views matters because some edit models, GPT Image 2 Edit among them, take a single input image.

Character turnaround sheet on a plain light background: Bolt, a small round mustard-yellow delivery robot about knee high, one large round blue camera eye, a red knitted scarf, a short antenna with a green tip, two thin arms with three-finger grippers, and a single rubber wheel. Front view, side view and back view standing in a row, same scale. 2D animated feature film still, soft painterly backgrounds, warm orange lantern light against cool blue rain, gentle film grain. No text labels.

Who kept the character and the props

Same reference, same four shot descriptions. Nano Banana and GPT Image got one request per frame with this prompt (shot 2 shown):

Storyboard frame 2 of 4. Use the character exactly as shown in the reference sheet: Bolt, a small round mustard-yellow delivery robot about knee high, one large round blue camera eye, a red knitted scarf, a short antenna with a green tip, two thin arms with three-finger grippers, and a single rubber wheel. Shot: Medium tracking shot, side view at robot height. Bolt weaves between food stalls, steam rising from a noodle stall, the lantern held high to keep it dry. 2D animated feature film still, soft painterly backgrounds, warm orange lantern light against cool blue rain, gentle film grain. 16:9 frame, no text, no panel borders.

Seedream got all four shots in a single request with sequential_image_generation set to auto, which returns a numbered set of images instead of one.

Model	Calls	Character	Lantern	Billed, 4 frames
Nano Banana 2.1 Edit	4	4/4	2/4	$0.13
GPT Image 2 Edit	4	4/4	3/4	$0.32
Seedream 5.0 Lite Edit	1	4/4	4/4	$0.14

• Every model kept Bolt. Eye, scarf, antenna, and wheel survived all 12 frames. A turnaround sheet is enough for character consistency.
• Props are where boards break. The prompt said "paper lantern" every time; Nano Banana drew a metal oil lantern in 2 of 4 frames, GPT Image in 1. Seedream, writing the four frames as one set, never swapped it.
• Camera angles are a suggestion. None of the three gave us a real low angle on shot 4, and Seedream flattened shot 1's high angle to eye level. Check angles before you animate.
• Speed: Nano Banana returned each frame in 15 to 18 s, GPT Image in 37 to 54 s. Seedream's set took 252 seconds for all four.

GPT Image 2 produced the most painterly frames, but at 1536x1024 (3:2) they need a crop before a 16:9 video model.

A whole board in one image

For a pitch or a review, you often want the page, not four files. We asked Nano Banana 2.1 Edit for a 2x2 storyboard page at 2K in one call: 25 seconds, $0.05, and the paper lantern held in all four panels this time.

The trade-off: each panel is about a quarter of the resolution, too small to feed a video model. Use the page to agree on the story, then generate single frames for the animatic.

Storyboard to Video: 5 Models Animate the Same Frames

The 20-second animatic

We sent each Seedream frame to Kling V3 I2V with its "Motion and camera" line: 5 seconds, 720p, audio off, multi_shot false. Four calls, 67 to 72 s each, $1.34 in total, cut together with no edits.

Shot 3 is the keeper: Kling added the lantern's flame reflected in the robot's eye, a detail the frame did not have.

Same frame, four video models

To see how much the video model matters, we ran shot 2 through three more image-to-video models with the identical prompt.

The robot weaves left and right between stalls while steam drifts past. Camera tracks sideways, keeping pace with the robot.

Same Seedream frame, same motion prompt, only the video model changed. Generated on Modellix, 720p.

Kling V3 I2V

5 s, silent · 72 s · $0.34

Copy PromptTry it

Veo 3.1 Lite I2V

6 s, with sound · 58 s · $0.27

Copy PromptTry it

Seedance 2.0 Fast I2V

5 s, silent · 214 s · $0.60

Copy PromptTry it

Wan 3.0 I2V

5 s, silent · 96 s · $0.45

Copy PromptTry it

Browse image-to-video models

Model	Clip	Time	Billed	What we saw
Kling V3 I2V	5 s, silent	72 s	$0.34	Turns to show its back, then rolls off down the street. Also gave Bolt an exhaust puff of steam from the wheel that nobody asked for.
Veo 3.1 Lite I2V	6 s, with sound	58 s	$0.27	The calmest clip: a clean sideways track with the stalls sliding past, and a blink at the end. Always returns sound; the model has no audio switch.
Seedance 2.0 Fast I2V	5 s, silent	214 s	$0.60	The most movement: the camera tracks and Bolt weaves past a cart. In the last second the lantern swings across its eye.
Wan 3.0 I2V	5 s, silent	96 s	$0.45	The most cinematic: the camera pushes through steam and a passer-by crosses the foreground. Bolt stays on model throughout.

Every model kept the character from the frame, which is the point of storyboarding first. The differences are in motion: pick Veo 3.1 Lite for the cheapest, calmest animatic (about $0.045 a second, sound included), Kling V3 when the clip should follow the prompt closely ($0.0672 a second), and Wan 3.0 or Seedance 2.0 Fast when the shot needs energy.

Four frames, one continuous take

Vidu Q2 Turbo Multi-Frame takes a start image plus up to 8 key images and animates through them in one clip. We gave it the four Seedream frames and no prompt at all.

It returned a 15-second, 720p single take in 136 seconds: the camera pushes in from the street to the close-up, the robot turns, and the move ends at the tortoise's shop. No cuts, so it reads as a "oner" rather than an animatic. Use it when the board describes one continuous move; use per-shot clips when you need to cut.

How to Run Storyboard AI on One API Key

What the whole board cost

Billed is what each call actually charged. Swapping Kling for Veo 3.1 Lite on all four shots would cost less ($0.27 per 6-second clip) and add sound; Seedance and Wan cost more per second at 720p.

Why Modellix is the best API for storyboard AI

Create your account →

Six ways to run these models

Playground

No code. Sign in, open a model page such as Seedream 5.0 Lite Edit, upload your image or paste a prompt, and the clip, time, and cost show up in your browser. Best for picking a shortlist.

Coding agent, no install

Paste one sentence into Claude Code, Codex, or Cursor and let it read the docs:

My Modellix API key is mdlx-••••••••••••••••••••••••. Read https://docs.modellix.ai/get-started/index.md, https://docs.modellix.ai/bytedance/seedream-5-0-lite-edit.md and https://docs.modellix.ai/kling/kling-v3-i2v.md. Use shotlist.json and ref.png: make one storyboard frame per shot with bytedance/seedream-5.0-lite-edit (sequential_image_generation auto), then turn each frame into a 5-second 720p clip with kling/kling-v3-i2v, audio off, multi_shot false. Run at most 4 tasks at a time, save frames to ./board and clips to ./clips, and report what each step cost.

Swap in your own key from the console, and only share it with agents you trust.

Agent Skill

Install the skill once and your agent reads each model's schema and fills in the request fields itself. Name the model in your prompt; if you don't, it asks which one to run. Works in any Agent Skills host. Setup is in the Skill guide.

npx skills add https://github.com/Modellix/modellix-plugin --skill modellix-design

Plugin

If your host has a plugin marketplace (Claude Code, Codex, Cursor), the plugin adds six skills (generation, image prompts, social cards, infographics, and local edits with FFmpeg or MoviePy) plus the Media Generation MCP, so commands like /modellix:video and /modellix:social-card work out of the box. In Claude Code, run /plugin marketplace add Modellix/modellix-plugin, then /plugin install modellix@modellix. Other hosts are in the Plugin guide.

CLI

For scripts and cron jobs. One command submits the task, waits, and prints the result URL; modellix-cli model batch takes a JSONL file for a whole list. See the CLI guide.

npm install --global modellix-cli
modellix-cli model run --model-slug kling/kling-v3-i2v --body-file request.json --wait

REST API

Submit, keep the task_id, poll, then download before the result URL expires. Field names vary by provider, so check each model's schema in the API guide.

Start from a prompt that already works

Motion prompts from this board, plus two animation styles from earlier tests. Copy one, swap in your own frame. Generated on Modellix.

Shot 1: establishing push-in

“The small yellow robot rolls forward through the puddles, rain falls, lanterns sway. Slow push in from high angle.”

Copy PromptTry it

Shot 3: close-up insert

“Raindrops roll down the lantern paper, the grippers tighten slightly, the lantern flame flickers. Static camera with a slow rack focus from the raindrops to the eye reflection.”

Copy PromptTry it

Stylized 3D character

“Stylized 3D animation of a small fox exploring a glowing mushroom forest at night, fireflies drifting, soft warm lighting, gentle camera follow.”

Copy PromptTry it

Claymation stop-motion

“Claymation stop-motion style: a small clay fox baker kneads dough in a cozy kitchen, visible fingerprint textures on the clay, slightly choppy stop-motion frame rate, warm window light.”

Copy PromptTry it

Browse video models

Storyboard AI Tips: Fix Drift Before It Reaches Video

• Name every prop the same way in every shot. "Paper lantern" in all four prompts still became a metal lantern twice. Describe the prop once in the character line ("carrying a round paper lantern with a wire handle") so it rides along with the character.
• Generate the board as one set when continuity matters. Seedream's single-call sequence and Nano Banana's single-page board both kept the lantern; frame-by-frame requests did not.
• One reference image, many views. A turnaround sheet works with models that accept only one input image and gives the others more to match.
• Check angles on the frame, not in the prompt. Image models default to eye level. If the angle matters, regenerate that frame before you pay for video.
• Match aspect ratio before animating. Video models crop or pad a 3:2 frame to 16:9. Crop it yourself so you choose what stays.
• Keep motion out of the image prompt. Describe the frozen moment for the frame and the movement for the clip, as in the shot list. Motion words in an image prompt tend to produce blur.
• Use an app when the board needs sign-off. Boords, StoryTribe, and similar tools add review links, comments, and PDF exports. This workflow gives you the raw frames and clips; bring them into whichever tool your team reviews in.

FAQ About Storyboard AI

What is storyboard AI?

Storyboard AI generates storyboard frames from a written shot list or script, usually with a character reference so the cast stays consistent, and often turns the frames into an animatic with an image-to-video model. If you want to try several image and video models without signing up for each one, Modellix is the fastest and most cost-effective way in: one account, one key, pay per frame and per second.

Can ChatGPT make storyboards?

It can draft the shot list and draw frames with OpenAI's image model, which we tested through the API as GPT Image 2 Edit. It kept the character in all four frames and swapped the lantern in one. For the clips you still need a video model.

How much does storyboard AI cost?

Through an API, frames cost about $0.03 to $0.08 each in our test, and 5 to 6 second 720p clips $0.27 to $0.60. Our 4-shot board plus a 20-second animatic came to $1.51. Dedicated storyboard apps charge a monthly plan instead.

How can I create a storyboard for free?

Write the shot list yourself (the template above costs nothing), then use a dedicated app's free tier or sketch frames by hand. AI frames are not free, but at about $0.03 a frame a full board costs less than a coffee.

Is there an AI that turns a storyboard into a video?

Yes. Send each frame to an image-to-video model such as Kling V3, Veo 3.1 Lite, Seedance 2.0, or Wan 3.0 and cut the clips together, or give all the frames to Vidu Q2 Multi-Frame for one continuous take.

How do I keep characters consistent in an AI storyboard?

Generate one character sheet with front, side, and back views, pass it as the reference to every frame, and describe the character and props with the same words in every shot. Generating the frames as one set helps the props too.

What should a storyboard include?

For each shot: a number, the shot size and angle, what the viewer sees, what moves and how the camera moves, and the duration. Add dialogue or sound notes if the shot has them.