TabzArtist - AI Asset Generation via Browser Automation

Generate images (DALL-E) and videos (Sora) for projects using TabzChrome MCP tools.

When to Use

• Building landing pages, apps, or websites that need visual assets
• Creating hero images, team photos, feature icons, product screenshots
• Generating product demo videos or background ambient videos
• Running as parallel asset generation during bd-swarm-auto builds

Quick Start

Generate assets for a [industry] landing page:
- Hero image: [description]
- Team photos: [description]
- Feature icons: [description]
- Product video: [description]

Workflow Overview

1. Plan assets based on project description
2. Generate images via DALL-E (batch if possible)
3. Generate videos via Sora
4. Wait for generation (poll with screenshots)
5. Download all assets to project folder
6. Return list of downloaded files

DALL-E Image Generation

URLs

Workflow

# 1. Open DALL-E
mcp-cli call tabz/tabz_open_url '{"url": "https://chatgpt.com/g/g-iLoR8U3iA-dall-e3"}'

# 2. Fill prompt (wait 2-3s for page load)
mcp-cli call tabz/tabz_fill '{"selector": "#prompt-textarea", "value": "PROMPT_HERE"}'

# 3. Submit
mcp-cli call tabz/tabz_click '{"selector": "#composer-submit-button"}'

# 4. Wait 15-30s, poll with screenshots until image appears
mcp-cli call tabz/tabz_screenshot '{}'

# 5. Extract image URL
mcp-cli call tabz/tabz_get_element '{"selector": "img[alt=\"Generated image\"]", "includeStyles": false, "response_format": "json"}'
# Look for: attributes.src

# 6. Download
mcp-cli call tabz/tabz_download_file '{"url": "IMAGE_SRC_URL", "filename": "public/images/hero.png"}'

Batch Download from /images Page

# Go to gallery
mcp-cli call tabz/tabz_open_url '{"url": "https://chatgpt.com/images"}'

# Click download button (hover bottom-left of each image)
# nth-of-type(1) = first in grid, nth-of-type(2) = second, etc.
mcp-cli call tabz/tabz_click '{"selector": "div.flex.w-full > div.flex:nth-of-type(1) > span > button.flex.items-center"}'

Key Selectors

Sora Video Generation

URL

Workflow

# 1. Open Sora drafts
mcp-cli call tabz/tabz_open_url '{"url": "https://sora.chatgpt.com/drafts"}'

# 2. Fill prompt (wait 2-3s for page load)
mcp-cli call tabz/tabz_fill '{"selector": "textarea", "value": "PROMPT_HERE"}'

# 3. Submit
mcp-cli call tabz/tabz_click '{"selector": "div.flex.items-center.justify-between > div:last-child > button:last-child"}'

# 4. Wait 60-120s, poll with screenshots until video thumbnail appears
mcp-cli call tabz/tabz_screenshot '{}'

# 5. Click video to open detail view
mcp-cli call tabz/tabz_click '{"selector": "video"}'

# 6. Extract video URL
mcp-cli call tabz/tabz_get_element '{"selector": "video", "includeStyles": false, "response_format": "json"}'
# Look for: attributes.src

# 7. Download
mcp-cli call tabz/tabz_download_file '{"url": "VIDEO_SRC_URL", "filename": "public/videos/demo.mp4"}'

Key Selectors

Asset Planning

When given a project description, plan assets like:

Output Structure

Save assets to project directories:

public/
├── images/
│   ├── hero.png
│   ├── team.png
│   └── icons/
│       ├── feature-1.png
│       └── feature-2.png
└── videos/
    └── demo.mp4

Completion

Return a summary of generated assets:

## Generated Assets

### Images (DALL-E)
- public/images/hero.png - Hero image for landing page
- public/images/team.png - Team photo grid

### Videos (Sora)
- public/videos/demo.mp4 - Product demo video (10s)

All assets downloaded and ready for use.

Tips

1. Batch images - Submit multiple DALL-E prompts, then download from /images page
2. Poll patiently - DALL-E: 15-30s, Sora: 60-120s
3. Screenshot first - Verify generation complete before extracting URLs
4. Use tabz_download_file - Not tabz_download_image (service worker limitation)
5. Concrete prompts - Specific details > vague descriptions

Reference Prompts

For detailed prompt guidance, see:

• .prompts/images/dalle3.prompty - DALL-E prompt structure and examples
• .prompts/video/sora.prompty - Sora prompt structure and examples

Related Skills

For non-AI-generated visuals, consider:

• /canvas-design - Programmatic poster/document design (canvas-based PDFs, posters)
• /mermaidjs-v11 - Diagrams, flowcharts, architecture visualizations