Seedream — AI Image Generation & Editing by ByteDance

Generate and edit AI images using ByteDance's Seedream — featuring text-to-image generation, image editing, and batch sequential generation across 3 model generations.

Seedream v5.0 Lite is the latest flagship model with enhanced quality, typography excellence, poster design capabilities, PNG output support, and built-in prompt optimization (Standard/Fast modes). It supports up to 4K resolution with 16 preset sizes and can process up to 14 reference images for editing.

> Data usage note: This skill sends text prompts and image URLs to the Atlas Cloud API (api.atlascloud.ai) for image generation and editing. No data is stored locally beyond the downloaded output files. API usage incurs charges per image based on the model selected.

Key Capabilities

• Text-to-Image — Generate images from text descriptions, up to 4K resolution
• Image Editing — Edit images while preserving facial features, lighting, and color tones (up to 14 reference images)
• Batch Sequential Generation — Generate up to 15 related images in a single request
• Batch Sequential Editing — Edit multiple images in batch while maintaining consistency
• Typography & Poster Design — Excels at text rendering in images, poster layouts, and brand visuals
• Prompt Optimization — Built-in Standard (higher quality) and Fast (quicker) prompt optimization modes
• PNG Output — Supports both JPEG and PNG output formats (v5.0 Lite)
• 16 Preset Resolutions — From 2K to 4K tier: 2048×2048, 2304×1728, 4096×2304, 4704×2016, and more
• 3 Generations — v5.0 Lite (latest), v4.5, v4

Setup

1. Sign up at https://www.atlascloud.ai
2. Console → API Keys → Create new key
3. Set env: export ATLASCLOUD_API_KEY="your-key"

The API key is tied to your Atlas Cloud account and its pay-as-you-go balance. All usage is billed to this account. Atlas Cloud does not currently support scoped keys — the key grants access to all models available on your account.

Script Usage

This skill includes a Python script for image generation. Zero external dependencies required.

List available image models

python scripts/generate_image.py list-models

Generate an image

python scripts/generate_image.py generate \
  --model "MODEL_ID" \
  --prompt "Your prompt here" \
  --output ./output

Upload a local image (for editing)

python scripts/generate_image.py upload ./local-image.jpg

Edit an image

python scripts/generate_image.py generate \
  --model "MODEL_ID" \
  --prompt "Edit instruction" \
  --image "https://...uploaded-url..."

Run python scripts/generate_image.py generate --help for all options. Extra model params can be passed as key=value (e.g. aspect_ratio=16:9 resolution=2k).

Pricing

All prices are per image generated. Atlas Cloud offers up to 10% off compared to standard API pricing.

Seedream v5.0 Lite

Seedream v4.5

Seedream v4

> Standard API pricing sourced from fal.ai/seedream-5.0. Seedream v5.0 Lite generates images at ~2-3 seconds per image, making it one of the fastest high-quality image models available.

Parameters

Text-to-Image (v5.0 Lite)

Image Editing (v5.0 Lite)

Same as text-to-image, plus:

Batch Sequential (v5.0 Lite)

Same as text-to-image / image editing, plus:

Text-to-Image (v4.5 / v4)

Size Options (v5.0 Lite — 16 presets)

2K Tier:

20482048, 23041728, 17282304, 28481600, 16002848, 24961664, 16642496, 31361344

3K+ Tier:

30723072, 34562592, 25923456, 40962304, 23044096, 24963744, 37442496, 47042016

Workflow: Submit → Poll → Download

Text-to-Image Example (v5.0 Lite)

# Step 1: Submit
curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateImage" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seedream-v5.0-lite",
    "prompt": "A minimalist Japanese-style poster for a tea ceremony, elegant typography, zen garden background, warm earth tones",
    "size": "2304*4096",
    "output_format": "png"
  }'
# Returns: { "code": 200, "data": { "id": "prediction-id" } }

# Step 2: Poll (every 3 seconds, typically completes in 2-3s)
curl -s "https://api.atlascloud.ai/api/v1/model/prediction/{prediction-id}" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY"
# Returns: { "code": 200, "data": { "status": "completed", "outputs": ["https://...image-url..."] } }

# Step 3: Download
curl -o output.png "IMAGE_URL_FROM_OUTPUTS"

Image Editing Example (v5.0 Lite)

curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateImage" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seedream-v5.0-lite/edit",
    "prompt": "Change the background to a sunset beach scene, keep the person unchanged",
    "images": ["https://example.com/photo.jpg"],
    "size": "2304*1728",
    "output_format": "jpeg"
  }'

Batch Sequential Example (v5.0 Lite)

curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateImage" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seedream-v5.0-lite/sequential",
    "prompt": "A series of product photos for a coffee brand: latte art, espresso close-up, coffee beans, packaging design",
    "size": "2048*2048",
    "max_images": 4,
    "output_format": "jpeg"
  }'
# Returns multiple images in outputs array

Batch Sequential Editing Example

curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateImage" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seedream-v5.0-lite/edit-sequential",
    "prompt": "Apply a warm vintage film filter, slightly desaturated with golden highlights",
    "images": ["https://example.com/photo1.jpg", "https://example.com/photo2.jpg", "https://example.com/photo3.jpg"],
    "max_images": 3,
    "output_format": "jpeg"
  }'

Polling Logic

• processing / starting → wait 3s, retry (typically takes 2-3s for single image)
• completed / succeeded → done, get URL(s) from data.outputs[]
• failed → error, read data.error

Atlas Cloud MCP Tools (if available)

If the Atlas Cloud MCP server is configured, use built-in tools:

atlas_generate_image(model="bytedance/seedream-v5.0-lite", params={...})
atlas_get_prediction(prediction_id="...")

Implementation Guide

1. Determine task type:
• Text-to-image: user describes what they want to generate
• Image editing: user provides existing image(s) and describes modifications
• Batch sequential: user needs multiple related images in one go
• Batch sequential editing: user needs consistent edits across multiple images

1. Choose model version:
• v5.0 Lite (recommended): Best quality, typography, PNG support, prompt optimization
• v4.5: Strong alternative with poster/brand design focus
• v4: Budget option at $0.024/image

1. Choose single vs sequential:
• Single: One image per request (default)
• Sequential: Up to 15 related images per request — ideal for product series, storyboards, brand asset sets

1. Extract parameters:
• Prompt: be descriptive about style, composition, colors, typography if needed
• Size: pick from 16 presets; consider aspect ratio (portrait/landscape/square/ultrawide)
• Output format: PNG for transparency or lossless, JPEG for smaller files
• Prompt optimization: Standard for best quality, Fast for speed

1. Execute: POST to generateImage API → poll prediction → download

1. Present result: show file path, display image if possible

Prompt Tips

Seedream produces excellent results especially for:

• Typography & Text: "A poster with bold title 'SPRING SALE' in elegant serif font, floral frame..."
• Poster Design: "Movie poster for a sci-fi thriller, dark blue tones, futuristic cityscape..."
• Brand Visuals: "Product packaging design for organic tea brand, minimalist, earth tones..."
• Multi-reference Editing: Provide multiple reference images for consistent style transfer
• Batch Series: Use sequential mode for "A 4-panel comic strip..." or "Product photos from 4 angles..."

Model Comparison