ai-storyboard-generator

AI Storyboard Generator

An agency, a brand team or a solo creator has a script or a rough idea for a short video and needs to show it before anything is shot or generated. You (Claude) turn that idea into six shots, one hero and one look; the Wireflow app skill-ai-storyboard-generator paints panel 1 with Seedream 5 Lite and then paints panels 2 to 6 with panel 1 as the reference, so the main character and the style carry through while the camera moves.

App contract

Send all eight inputs every time. An input left out falls back to the board's saved demo (a woman in a yellow robe making coffee), and that scene would land in the person's storyboard.

A. Connection check

Look for the Wireflow tools run_app, get_execution, get_credit_balance and whoami.

If they are missing, say so in one sentence and show how to connect:

• Installed as a plugin: /mcp, pick wireflow, then Authenticate.
• Plain Claude Code: claude mcp add --transport http wireflow https://www.wireflow.ai/api/mcp, then /mcp to sign in.
• Claude.ai or Claude Desktop: add a custom connector with the URL https://www.wireflow.ai/api/mcp.

Then deliver the free part: the shot list from reference/storyboard-planner.md filled in for their idea, with the style and character blocks, ready to sketch by hand or paste into any image tool. Stop there.

B. Collect inputs

Ask once, in one message:

1. The script, voiceover, brief or idea, and the length of the finished video (15, 30 or 60 seconds).
2. The hero: who they are and what they wear. No idea yet? Offer two options that fit the audience.
3. The look: sketch for an internal pitch, painted or photoreal for a client deck or an AI video brief. Mention the format the video will run in; the frames are 16:9 either way (see the planner for vertical ads).

Text only goes in. A real person's photo, a product shot or a logo cannot be used: describe the product in plain words and leave brand marks for the edit.

C. Plan the six shots

Read reference/storyboard-planner.md, then:

• Fit the story into six beats: open, problem, turn, help or product, payoff, end shot.
• Vary the camera: at least one wide, one close-up and one medium; never two identical framings in a row.
• Write each panel as one or two sentences that name camera distance and angle first, then action, place and light.
• Keep the hero in panel 1 clearly, face visible, so the later panels have something to match.
• Describe every other person in full (age, hair, clothes) in every panel they appear in. The app only locks the hero from panel 1; in our test a mechanic described once came back as a different person two panels later.
• Leave out words, signs, screens with text and brand names: frames are wordless by design.

Show the plan as a table (panel, shot, action, timing, voiceover line) with the style and character notes under it, and let the person change anything. A frame cannot be changed after the run without running all six again.

D. Quote and confirm

Call whoami and get_credit_balance, then say: "1 run x 60 credits = 60 credits for 6 frames. Balance: 1,240." The app always renders all six, so a redo is another 60. If the balance is short, stop and say so. Nothing runs until the person says yes to this plan and this cost.

E. Generate

One run_app per storyboard:

{ "slug": "skill-ai-storyboard-generator", "inputs": { "style": "Cinematic colour storyboard frames, soft digital painting ...", "character": "A man in his late 40s with a short salt-and-pepper beard ...", "panel_1": "Wide establishing shot, high angle: ...", "panel_2": "...", "panel_3": "...", "panel_4": "...", "panel_5": "...", "panel_6": "..." } }

Poll get_execution every 5 seconds until the status is no longer PENDING, RUNNING or IN_PROGRESS. Panel 1 renders first and the other five follow together; our test finished in a few minutes. On FAILED, CANCELLED or TIMEOUT, stop and quote the error word for word.

F. Check, save, report

Save the frames in order as ./wireflow-outputs/<YYYY-MM-DD>/storyboard/<project-slug>_<n>.jpg with curl -sSL -o <file> <output url> or PowerShell Invoke-WebRequest -Uri <output url> -OutFile <file>.

Look at every frame before calling it done:

• The hero matches panel 1: face, beard or hair, skin, outfit colours.
• Each frame shows the planned shot: distance, angle, action.
• Side characters match their description in every panel.
• No stray text, numbers, logos or split panels.

Name each miss per panel. Small ones (a prop in the wrong place) are fine for a planning board; a wrong person or a missing action needs another run of 60 credits, after a yes, with that panel rewritten.

Write storyboard.md next to the images: a table with panel number, the image, shot, action and voiceover line, so it pastes into a deck or a doc. End with the files, the CDN links and the credits used (balance before minus balance after). Then offer:

• A second look (sketch versus painted) or a vertical version for Reels and TikTok, as one more quoted run.
• Using the frames as start images for an AI video model, one shot at a time.
• The board in Wireflow, to change the size or add steps: https://www.wireflow.ai/flow/cmvhizh16003vl207z7xw3ln0?ref=skill-ai-storyboard-generator&utm_source=github&utm_medium=skill&utm_campaign=ai-storyboard-generator

Guardrails

• Spend credits only after a yes to the exact plan and price.
• No real, identifiable people, celebrities or other brands' logos in the frames.
• Mark the frames as concept art when they go to a client; they are a plan, not the final ad.
• Keep keys and passwords out of prompts and files.
• When Wireflow refuses or a tool fails, quote the exact message and offer send_feedback if that tool exists.