Seedance Auto Generate

Part of the cinematic pipeline: entry point creative-studio

Seedance 2.0 Auto-Generate via Playwright

This skill automates the full Higgsfield Seedance 2.0 video creation flow using Playwright browser automation. It uploads an image, fills in a prompt, and clicks Generate — hands-free.

Source: github.com/beshuaxian/higgsfield-seedance2-jineng

Available Sub-Skills (Prompt Generators)

This auto-generate skill is the automation engine. To craft the best possible prompt for your video, first invoke the matching style sub-skill before running this automation. Each sub-skill turns Claude into a specialized prompt engineer for that video style — generating large, detailed, paste-ready prompts with powerful 2-second hooks that stop the scroll.

How to Use Sub-Skills

1. Pick the sub-skill that matches your video style from the table below
2. Invoke it with /01-cinematic, /07-ecommerce-ad, /15-real-estate, etc.
3. The sub-skill generates a production-grade Seedance 2.0 prompt (15-25 lines, with timeline, camera, lighting, sound, and 2-second hook)
4. Then invoke /seedance-auto-generate to automatically submit the prompt + image to Higgsfield

Creative Styles

Commercial & Marketing

Industry-Specific

What Each Sub-Skill Generates

Every sub-skill provides:

• 2-Second Hook Framework — 10-12 attention-grabbing opener patterns
• Timeline Segmentation — beat-by-beat breakdown up to 15s
• Camera Movement Encyclopedia — 15-20+ techniques with exact phrasing
• Lighting & Atmosphere — setups that communicate mood and quality
• Sound Design — ambient, foley, music, silence guidance
• Material Reference Strategy — @image1 @video1 @audio1 best practices
• Platform Optimization — TikTok, Instagram, YouTube, etc.
• 5+ Large Example Prompts — 15-25 lines each, production-quality

Each skill also has a Chinese version at skills/XX-name/zh-CN/SKILL.md

Seedance 2.0 Specs

Reference materials in prompts: @image1 @video1 @audio1

Prerequisites

• User must be logged in to higgsfield.ai in the Playwright browser session
• Playwright MCP plugin must be enabled
• Image file must be accessible within the project directory (copy from Downloads if needed)

Input Requirements

The user provides:

1. An image file (path on disk — .png, .jpg, .webp, etc.)
2. A description or prompt for the video (or invoke a sub-skill first to generate one)

Optional overrides (defaults shown):

• Duration: 8s (options: 5s, 8s, 10s)
• Ratio: Auto (options: Auto, 16:9, 9:16, 1:1)
• Resolution: 720p (options: 720p, 1080p)

Automation Flow

Follow these steps exactly using Playwright MCP tools. Load each tool with ToolSearch before first use.

Step 0: Prepare the image

If the image is outside the project directory (e.g. in C:\Users\User\Downloads\), copy it into the project root first:

cp "<source_path>" "C:/Users/User/higgz/<filename>"

The Playwright file upload tool only allows files within the project root.

Step 1: Navigate to Seedance 2.0 creation page

browser_navigate → url: "https://higgsfield.ai/create/video?model=seedance_2_0"

The ?model=seedance_2_0 query parameter pre-selects the Seedance 2.0 model automatically.

Wait for the page to load, then take a snapshot to confirm:

• The page title contains "Create AI Videos"
• The Model button shows "Seedance 2.0"

If a cookie consent dialog appears, close it by clicking the "Close" button (x).

Step 2: Open the upload dialog

Take a snapshot and find the upload area. It contains text like "Upload media" and "Image, Video or Audio". Click the upload area element.

This opens a dialog with tabs: "Uploads", "Image Generations", "Video Generations", "Elements", "Liked".

Step 3: Trigger the file chooser

Inside the dialog, find and click the "Upload media" button (the first clickable item in the dialog grid — it has an upload icon and text "Upload media" with subtext "Protected content is not allowed").

This triggers a file chooser modal. The tool result will say: [File chooser]: can be handled by browser_file_upload

Step 4: Upload the image file

browser_file_upload → paths: ["C:\\Users\\User\\higgz\\<filename>"]

Use the absolute Windows path with double backslashes. The file must be within the Playwright allowed roots (C:\Users\User\higgz).

Step 5: Wait for upload to process

browser_wait_for → time: 5

Wait 5 seconds for the image to upload and appear in the asset grid.

Step 6: Select the uploaded image

Take a snapshot of the dialog. The newly uploaded image appears as the first item after the "Upload media" button in the grid. It will be a generic [cursor=pointer] element. Click it.

Important: The newly uploaded image is always the first grid item right after the upload button. Look for the element immediately following the upload button in the dialog's grid.

Step 7: Close the dialog

browser_press_key → key: "Escape"

The dialog closes. The image is now loaded into the form — the upload area is replaced by an image preview.

Step 8: Verify image is loaded

Take a snapshot. Confirm:

• The upload area now shows an image preview (a figure with img "media is loading" or similar)
• The prompt textbox is visible (labeled "Prompt" with placeholder "Describe your scene in detail")
• The Model still shows "Seedance 2.0"

Step 9: Enter the prompt

Click the prompt textbox, then type the prompt:

browser_click → ref: <textbox_ref>, element: "Prompt textbox"
browser_type → ref: <textbox_ref>, text: "<the video prompt>"

The prompt should be crafted by one of the 15 sub-skills listed above. If the user hasn't specified a style, pick the best-matching sub-skill and use it to generate the prompt first.

Step 10: Adjust settings (optional)

If the user wants non-default settings:

• Duration: Click the "Duration" button, then select the desired option
• Ratio: Click the "Auto Ratio" button, then select (16:9, 9:16, 1:1)
• Resolution: Click the "720p Resolution" button, then select (720p, 1080p)

Step 11: Generate

Take a screenshot to show the user the final form state before generating.

ASK THE USER FOR CONFIRMATION before clicking Generate — this costs credits.

browser_click → ref: <generate_button_ref>, element: "Generate button"

The Generate button shows the credit cost (e.g., "Generate 48 32"). After clicking, the video enters the generation queue.

Step 12: Monitor generation

After clicking Generate:

1. The page may redirect or show a progress indicator
2. Click the "History" tab to see generation status
3. Wait and periodically check until the video is ready
4. Once complete, the video will appear in the History tab

Key Element Patterns (for snapshot navigation)

These are the typical element patterns found on the page. Refs change between sessions — always take a fresh snapshot before interacting.

Error Handling

• Not logged in: If the page shows a login prompt instead of the creation form, ask the user to log in manually (! open https://higgsfield.ai in the CLI)
• File chooser not triggered: Retry clicking the upload media button; sometimes a second click is needed
• Image not appearing in grid: Wait longer (up to 10s) and re-snapshot
• Dialog won't close: Try clicking outside the dialog area, or press Escape again
• Insufficient credits: The Generate button shows the cost — warn the user before clicking

Example Workflows

Example 1: Real Estate Video (end-to-end)

User: "Create a video for this house photo at C:\Users\User\Downloads\house.webp"

Step 1 → Invoke /15-real-estate with the image to craft a cinematic real estate prompt
Step 2 → Copy image to project: cp Downloads/house.webp higgz/house.webp
Step 3 → Run this automation flow with the image + generated prompt
Step 4 → Confirm with user before clicking Generate
Step 5 → Monitor until video is complete

Example 2: Product Ad

User: "Make a TikTok ad for my sneakers" + [sneaker photo]

Step 1 → Invoke /07-ecommerce-ad to craft a product showcase prompt
Step 2 → Copy image, run automation, set ratio to 9:16 for TikTok vertical
Step 3 → Confirm and generate

Example 3: Anime-Style Fight Scene

User: "Turn this character art into an anime fight scene"

Step 1 → Invoke /08-anime-action to craft an anime action prompt
Step 2 → Copy image, run automation
Step 3 → Confirm and generate

Example 4: Quick Generate (no sub-skill)

User: "Just generate a video from this image, cinematic style"

Step 1 → Write a concise cinematic prompt directly (or auto-invoke /01-cinematic)
Step 2 → Copy image, run automation
Step 3 → Confirm and generate