Prices shown are starting prices. Higher resolution or additional features may cost more.

Input Parameters

Text-to-Image (Generate Mode)

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | prompt | string | (required) | Text description of the desired image | | size | string | 1024x1024 | Output image dimensions (e.g., 512x512, 1024x1024, 1280x720) | | num_inference_steps | number | 25 | Number of denoising steps (higher = better quality, slower) | | seed | number | random | Random seed for reproducible results | | guidance_scale | number | 5.0 | How closely to follow the prompt (higher = more faithful) | | num_images | number | 1 | Number of images to generate (1 for standard, up to 15 for sequential) | | negative_prompt | string | — | What to avoid in the generated image |

Image Editing (Edit Mode)

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | prompt | string | (required) | Text description of the desired edit | | image | string | (required) | Base64-encoded image or URL of the source image | | mask_image | string | — | Base64-encoded mask or URL (white = edit area, black = preserve) | | strength | number | 0.75 | Edit strength (0.0 = no change, 1.0 = full regeneration) | | size | string | from source | Output image dimensions | | num_inference_steps | number | 25 | Number of denoising steps | | seed | number | random | Random seed for reproducible results | | guidance_scale | number | 5.0 | How closely to follow the prompt | | negative_prompt | string | — | What to avoid in the edit |

Sequential / Batch Mode

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | prompt | string | (required) | Text description for the batch | | num_images | number | (required) | Number of images to generate (up to 15) | | size | string | 1024x1024 | Output image dimensions | | num_inference_steps | number | 25 | Number of denoising steps | | seed | number | random | Base random seed | | guidance_scale | number | 5.0 | How closely to follow the prompt | | negative_prompt | string | — | What to avoid in the generated images |

Why Seedream?

Best Chinese Text Rendering

Seedream is the undisputed leader in rendering Chinese characters within generated images. Whether you're creating posters, banners, social media graphics, or product mockups with Chinese typography, Seedream produces crisp, accurate, and beautifully styled text that other models simply cannot match.

Poster & Design Excellence

The Seedream model family has been specifically optimized for commercial design use cases:

• Magazine covers with perfect text placement
• Product posters with accurate brand typography
• Social media graphics with legible overlay text
• Event banners with complex multi-line layouts
• Marketing materials with mixed Chinese/English text

NSFW Whitelisted on Atlas Cloud

Atlas Cloud provides whitelisted access to Seedream's full capabilities, including NSFW content generation. This is ideal for:

• Adult content platforms
• Artistic nude photography generation
• Unrestricted creative exploration
• Content that other platforms would block

Sequential Generation

Need a series of related images? Seedream's sequential mode generates up to 15 coherent, thematically linked images in a single API call — perfect for:

• Storyboard creation
• Product angle variations
• Style exploration
• A/B testing visual concepts
• Comic or manga panel generation

Installation

Prerequisites

• Bun runtime installed
• An Atlas Cloud API key (get one here)

Setup

# Clone the repository
git clone https://github.com/thoughtincode/seedream-image-skill.git
cd seedream-image-skill

# Install dependencies
bun install

# Set up your API key
cp .env.example .env
# Edit .env and add your ATLAS_CLOUD_API_KEY

# Link the CLI globally
bun link

Verify Installation

seedream --help

Usage

Basic Text-to-Image

seedream "A majestic dragon flying over a mountain range at sunset"

Specify Model and Size

seedream "Cyberpunk cityscape at night" --model v4.5 --size 1280x720

Image Editing

seedream "Add a rainbow in the sky" --mode edit --image ./photo.png --strength 0.6

Edit with Mask

seedream "Replace with a red sports car" --mode edit --image ./scene.png --mask ./mask.png

Batch Generation

seedream "Cute cat in different poses" --mode batch --count 5

Sequential with Custom Parameters

seedream "Comic panels of a space adventure" --mode batch --count 8 --size 768x768 --steps 30

NSFW Mode

seedream "Artistic figure study, oil painting style" --nsfw

Full Parameter Control

seedream "A beautiful Japanese garden in autumn" \
  --model v5.0-lite \
  --size 1024x1024 \
  --steps 30 \
  --guidance 7.5 \
  --seed 42 \
  --output ./my-garden.png \
  --negative-prompt "blurry, low quality"

CLI Reference

seedream <prompt> [options]

Arguments:
  prompt                    Text description of the image to generate

Options:
  --model <variant>         Model variant: v5.0-lite (default), v4.5, v4
  --mode <mode>             Generation mode: generate (default), edit, batch
  --size <WxH>              Image dimensions (default: 1024x1024)
  --steps <n>               Inference steps (default: 25)
  --guidance <n>            Guidance scale (default: 5.0)
  --count <n>               Number of images for batch mode (default: 1, max: 15)
  --seed <n>                Random seed for reproducibility
  --output <path>           Output file path (default: ./output/<timestamp>.png)
  --image <path>            Source image for edit mode (file path or URL)
  --mask <path>             Mask image for edit mode (file path or URL)
  --strength <n>            Edit strength 0.0-1.0 (default: 0.75)
  --negative-prompt <text>  What to avoid in the generation
  --nsfw                    Enable NSFW content generation
  --help                    Show this help message

Agent Skill Integration

This tool is designed to work as an AI agent skill across all major coding agents. Install it with a single command:

npx skills add seedream-image-skill

This works with Claude Code, Cursor, OpenAI Codex, GitHub Copilot, Gemini CLI, Windsurf, OpenCode, Kiro, and any agent that supports skill installation.

Add to your project's CLAUDE.md

## Image Generation with Seedream

Use the `seedream` CLI to generate images:
- Basic: `seedream "your prompt here"`
- Edit: `seedream "edit instructions" --mode edit --image ./source.png`
- Batch: `seedream "prompt" --mode batch --count 5`
- NSFW: add `--nsfw` flag
- Models: v5.0-lite (default), v4.5, v4

How It Works

1. Your AI agent reads the skill instructions
2. When asked to generate images, the agent invokes the seedream CLI
3. The CLI sends requests to Atlas Cloud API
4. Images are downloaded and saved to the ./output directory
5. The agent can reference the saved images in subsequent interactions

API Details

Authentication

Set your Atlas Cloud API key in the .env file or as an environment variable:

export ATLAS_CLOUD_API_KEY=your_api_key_here

Endpoints

| Endpoint | Method | Description | |----------|--------|-------------| | https://api.atlascloud.ai/api/v1/model/generateImage | POST | Submit image generation request | | `https