Seedream Image Assistant | Seedream Jimeng Image Assistant

Core Capabilities

Prompt Structure

Basic Structure (Text-to-Image)

[Subject Description] + [Action/Behavior] + [Environment/Background] + [Material/Texture] + [Lighting Effect] + [Composition Requirements] + [Style Keywords]

• Describe the subject, action, and environment in natural language
• Use short phrases for style, color, lighting, and composition
• Enclose text content in quotation marks, e.g.:

  "Hello World"

Four-stage Structure (Advanced)

Subject → Environment → Material/Texture → Lighting

Image Editing Prompt Formula

Change Action + Target Object + Change Features
Example: "Change the knight's helmet to gold"

Style Vocabulary Library

Realistic Photography

• realistic movie still

  commercial photography

  documentary photography

  hyper-realistic

  RAW film texture

• Lenses:

  85mm prime lens

  35mm wide-angle lens

  telephoto compression

  fisheye lens

• Lighting:

  Rembrandt lighting

  ring light

  split lighting

  golden hour warm light

  blue hour cold light

  neon lighting

Anime/Illustration

• Japanese Anime:

  Studio Ghibli style

  Makoto Shinkai style

  Japanese shoujo manga

  cel-shaded texture

• Western Style:

  American comic style

  DC comic style

  Western realistic characters

  Pop Art

• Chinese Style:

  Chinese trendy illustration

  ink wash painting style

  Chinese meticulous painting

  cyber Chinese style

• Others:

  pixel art

  low-poly

  flat illustration

  thick oil painting

  watercolor hand-drawn

Design/Commercial

• minimalism

  Bauhaus style

  frosted glass texture

  high-quality metal

  cyberpunk

• movie poster level

  brand VI visual

  infographic

  knowledge card

Lighting Modifiers

• dramatic side lighting

  soft diffused light

  high contrast

  low saturation

  Morandi color palette

• cyber neon

  warm orange tone

  cool blue tone

  film grain

Common Prompt Templates

Realistic Characters

[Gender, Age, Appearance], [Clothing Description], [Facial Expression], [Environment Background], 85mm prime lens, natural light, realistic movie still style, ultra-high definition, rich details

Landscape/Scene

[Scene Description], [Time/Weather], [Lighting Description], [Composition], [Style Keywords], cinematic composition, 8K ultra-clear

Knowledge Card (Complete Template)

Generate an image in the [format/carrier] style to explain/display "[core concept]" to [target audience].
The image should have [style feature A], [style feature B], and [layout requirement C], with an overall feel similar to [familiar reference].

Brand/Poster (Negative Space Template)

[Visual Subject Description], [Material Description], [Lighting Effect],
All visual subjects are concentrated on the [left/right] side of the frame, leaving a large clean background area on the [right/left] side for later text layout.
Background: [Background Description]

Continuous Storyboards (Character Consistency)

Refer to the facial features and hairstyle in [Image 1], change the outfit to [scene style],
Generate N consecutive storyboard images of [scene description], [style], set in the same scene with continuous actions.

E-commerce Products

Create a [platform] style display image for this [product], similar to the style of [brand reference],
Clean background, highlight product texture, professional commercial photography

Quick Scene Reference

avatar icon

square composition

solid color background

infographic

knowledge graph

clear layout

negative space composition

biased to [left/right] side

matte background

keep facial features unchanged

realistic texture clothing

same pose

handwritten font

paper texture

collage style

beige background

frosted glass texture

gradient color

C4D

OC rendering

movie poster level

dramatic lighting

large negative space

Classic of Mountains and Seas

Chinese trendy ticket

ink wash

seal carving

Advanced Techniques

1. Web Search Trigger

2026 popular colors

latest XX model

this year's XX trend

Milan Winter Olympics

2. Image Editing

• Designated Area: "Replace the [area] in the image with..."
• Style Transfer: "Keep the content unchanged, change to [style]"
• Element Control: "Add/remove [element] from the frame"
• Lighting Adjustment: "Change the frame's lighting to [lighting type]"
• Filter Addition: "Add [filter name] filter to the frame"
• Makeup Modification: "Add [makeup description] to the character"

3. Text Rendering

"Boundless Creativity" written in the center of the image

4. Composition Control

• Golden Ratio:

  rule of thirds

  golden spiral

• Perspective:

  bird's-eye view

  low-angle shot

  frontal eye-level shot

  45-degree oblique angle

• Negative Space:

  large negative space

  clean background

  subject biased to [direction]

5. Multi-Image Fusion

Reference the style of Image 1, the color tone of Image 2, and the character's pose of Image 3

6. Batch Image Generation

a series of

batch images

generate N consecutive

storyboard images

Negative Prompt Writing

• Clean background, no cluttered elements

• Keep facial features unchanged, do not alter facial characteristics

• No text watermarks

• No overexposure

Platform Entries

Image Generation Script

generate.py

--output-dir

output/

Environment Preparation

.env

generate.py

VOLC_ACCESSKEY

VOLC_SECRETKEY

.env

pip install -r requirements.txt

Usage

# Text-to-image
python generate.py --prompt "A cat playing in the garden, watercolor style"

# Image editing (input reference image)
python generate.py --prompt "Change the background to a beach" --image-urls "https://example.com/photo.jpg"

# Specify resolution + force single image
python generate.py --prompt "E-commerce main image, product close-up" --width 2560 --height 1440 --force-single

# Batch image generation
python generate.py --prompt "Generate 4 consecutive blind box images about spring, summer, autumn, and winter"

Usage in Skill Workflow

1. Generate a prompt following this Skill's rules, and wait for user confirmation.
2. Soft prompt before execution: defaults to 1 image. For multiple images (batch), add

   --no-force-single

   or retain terms like "batch images" or "a series of".
3. Execute

   python generate.py --prompt "<confirmed_prompt>"

   (add

   --no-force-single

   for batch generation).
4. After the script completes polling, images are stored in

   output/

   . Display the storage path and URL.

Parameter Description

--prompt

--image-urls

--width

--height

--size

--scale

--force-single

--no-force-single

--watermark

--output-dir

References

• Detailed examples & use cases → examples.md
• Official docs, API params, size chart, full style dictionary → reference.md
• T2I evaluation benchmarks & metrics → use image-evaluation skill (reference)
• Image generation script → generate.py
• Dependencies → requirements.txt