imagineart

name	imagineart
description	Drive ImagineArt via Playwright to generate a video from a prompt (and optional reference image). Use when the user asks to "generate a video", "make an ImagineArt shot", "create an ImagineArt clip", or names a shot they want rendered. Opens a visible Chrome window so the browser automation is on-camera.

ImagineArt — automated generator

This skill turns a creative brief into a finished video clip by driving ImagineArt's web UI directly. It's designed to be visible — a real Chrome window opens, Claude's script clicks through the UI, and the user watches the generation happen in real time. This is the shot used in the ImagineArt × Claude video.

When to invoke

• User says some variant of: "generate a video of X", "make an ImagineArt clip", "run it through ImagineArt", "create a shot".
• User provides a prompt, an optional reference image, or both.
• User is ready for a visible browser automation run (not silent/headless).

Special intent: "run the product shoot" / "full campaign" / "autopilot"

If the user says "run the product shoot", "full campaign", "autopilot", "build me a campaign", or anything that implies generating a complete set of hero video + style variants + upscale from one product photo, route to the product shoot flow:

cd "/Users/liamjohnston/Coding Projects/Higgsfield X Claude" && \
  node automation/product-shoot.js

Edit PRODUCT_IMAGE and PRODUCT_DESCRIPTION at the top of product-shoot.js to match the product. The script runs three stages automatically: hero video (Seedance 2, 360° orbit), three brand-style image variants, and a 4K upscale — all without touching anything.

Special intent: "create AI UGC" / "create a UGC creator"

If the user says "create AI UGC", "make a UGC creator", "generate a creator", "run the UGC factory", or anything that implies synthesizing a character and animating it through multiple shots, route to the UGC factory flow instead of the single-shot flow:

cd "/Users/liamjohnston/Coding Projects/Higgsfield X Claude" && \
  node automation/ugc.js

That script has the full character brief + shot library hardcoded at the top. It generates images then runs them through the video generator for each shot in the library. Edit the constants in ugc.js to change the character or shot list.

What to do

1. Gather the brief

If the user gave a vague brief ("make me a cool product shot"), do not invoke the automation yet. First, architect the prompt using the rules in prompt-architecture.md (read it now if you haven't). A good ImagineArt video prompt specifies:

• Subject — who/what, described visually
• Action — what they're doing, with specific verbs
• Camera motion — dolly in, orbit, handheld, static, crane
• Lighting — key light direction, color, hardness
• Style — cinematic, documentary, anime, etc.
• Duration/pacing cue — optional

Draft the prompt, show it to the user in a code block, and confirm before running. If the user already wrote a prompt, skip confirmation and use theirs verbatim.

2. Decide on a reference image

ImagineArt supports image-to-video. Options:

• User supplied an image path → use it
• User wants a fresh image → pause and ask if they want to generate one first or proceed text-only
• Check references/ folder — if there's an obvious match for the brief, offer it

3. Run the automation

Invoke from the project root:

cd "/Users/liamjohnston/Coding Projects/Higgsfield X Claude" && \
  (cd automation && npm install --silent 2>&1 | tail -5 && npx playwright install chromium --with-deps 2>&1 | tail -3) && \
  node automation/run.js \
    --prompt "FULL_PROMPT_HERE" \
    --image "references/IMAGE_FILE" \
    --out outputs

Notes:

• npm install and playwright install chromium are idempotent — safe to re-run, they no-op if already installed.
• Run without --image for text-only generation.
• Watch the terminal output. The script screenshots each step into outputs/ so you have visual evidence of every click.

4. First-run login

On the first run, ImagineArt will show the sign-in page. The script detects this and waits up to 5 minutes for the user to log in manually. Do not try to enter credentials yourself — credentials handling is a prohibited action. Tell the user out loud: "Log into ImagineArt in the window that just opened, then I'll take over."

Subsequent runs reuse the .chrome-profile/ directory and skip login.

5. Report the result

When node run.js exits, parse its === RESULT === JSON block. Surface:

• The video URL (videoUrl field)
• The screenshot paths (for the user to drop into their video edit)
• Any warnings (e.g., "video model option not found" — means selectors need patching)

If selectors failed, run node automation/recon.js next — it dumps the current DOM so you can patch SELECTORS in automation/imagineart.js.

Files in this skill

• SKILL.md — this file
• prompt-architecture.md — how to write a good ImagineArt video prompt (read before drafting)

Files in the project used by this skill

• automation/imagineart.js — Playwright driver (selectors + flow)
• automation/run.js — CLI wrapper
• automation/recon.js — DOM dump for patching selectors
• references/ — source images for image-to-video
• outputs/ — generated videos + step-by-step screenshots
• .chrome-profile/ — persistent Chromium profile (login state)

Failure modes and recovery

Symptom	Cause	Fix
"Could not find video model option" warning	ImagineArt renamed/moved the model picker	Run recon.js, patch SELECTORS.modelPickerTrigger and SELECTORS.videoModelOption in imagineart.js
Video never appears, 5min timeout	Generation queue slow, or result DOM changed	Check screenshots in outputs/. If generation finished visually but selector missed it, patch SELECTORS.resultVideo
Login loop — script keeps waiting	Cloudflare or 2FA blocking	User finishes challenge manually; the wait-for-login loop handles it
Prompt typed into wrong element	Multiple textareas/contenteditables on page	Tighten SELECTORS.promptTextarea using recon output