Lumina Documentation

What is Canvas?

Canvas is an all-in-one creative workspace where users can bring together text, images, video, audio, and music in one place. It helps with layout, editing, asset organization, and exporting, so teams can turn ideas into polished content more easily.

Instead of relying on traditional manual design workflows, Canvas makes creative production faster and more flexible. It supports users with different skill levels and helps them build content in a more intuitive way.

From creative ideation to asset assembly, layout design, and final output, Canvas covers the full workflow.

In short: Canvas helps users create faster, present ideas more professionally, and produce content that is more compelling across different scenarios.

Unlimited Workflows, Unlimited Creativity

Start with an idea and build freely on an infinite canvas. Combine different capabilities, branch into multiple creative directions, and stay in the flow as you create.

Built on In-House Innovation

Canvas brings together the latest and most comprehensive in-house creative capabilities, along with Seed’s leading AIGC vision models, to support stronger and more reliable output.

Real-Time Preview, Step-by-Step Editing

Run any part of a workflow with precision and see results instantly. With real-time feedback, users can adjust and refine the creative flow at any stage.

Easy Sharing, Fast Reuse

Workflows can be saved as personal assets or shared across teams for collaboration and reuse. With one-click deployment, teams can move faster and work more efficiently.

Agent makes Canvas truly intelligent

With Canvas, users can build workflows by connecting nodes and components. This already makes it possible to deliver creative outputs at scale with greater consistency and stability.

But in real projects, users often run into bigger questions:

• Can workflow building be made simpler?

• Can one canvas automatically choose the right path for different scenarios?

• Can it generate strong prompts more consistently, without jumping between multiple tools?

• Can a single canvas support the full pipeline from script → storyboard → video?

This is where Agent becomes powerful.

Think of Agent as the control center of Canvas.

It can understand natural language instructions, decide what needs to be done, and accurately call the right tools—or even entire workflows—to complete the task.

In other words, Agent turns Canvas from a manual workflow builder into a more intelligent, automated creative system.

Canvas Key Concepts

Basic

Term	Description	Preview
Node	The smallest building block of a workflow. Each node performs a specific function, and by combining different nodes, users can create a wide range of creative outcomes.
Connection	Connections link nodes together, allowing data to flow between them. This enables users to build complex workflows—similar to assembling building blocks.
Model	Models power the output of a workflow. By selecting different models, users can generate different styles, qualities, and types of content. Lumina currently supports ByteDance’s in-house image and video models, with more high-quality models to be added over time.

Advanced

Term	Description	Preview
Component (official)	Components allow multiple nodes to be grouped into a reusable unit. By defining inputs, outputs, and parameters, components simplify workflow design and improve reusability.
Agent	Agent acts as the control center of Canvas. It understands user instructions and automatically selects and runs the right tools (components), or even entire workflows, to complete a task.
AI apps	After testing, workflows can be published and reused across teams. - One-click deployment: Publish directly from Canvas - Auto-loaded parameters: Reuse pre-configured settings - Quick setup: Simplified configuration for faster adoption

How to Create a Canvas?

Step	Instruction	Preview
Step 1: Go to Canvas	Click Canvas in the left-side navigation bar. Note: Canvas is available only to Standard, Advanced, and Ultra subscribers.
Step 2: Create a Canvas	There are two ways to get started: 1. Start from a workflow template Click Workflow Template at the top, then click Create Now to open the template workflow and start editing.
	2. Start from scratch Click Create Canvas in the top-right corner to build a brand-new workflow from the ground up.

Canvas Quick Actions

Action	Description	Preview
Add nodes	Click the “+” button in the bottom toolbar, or double-click anywhere on the blank canvas.
Upload assets	Double-click the canvas, choose File, and upload your asset. You can also search for nodes such as “lora image,” “load video,” or “load audio” based on your needs.
Connect nodes	After adding nodes to the canvas, drag outward from an output port with the left mouse button to create a connection.
Delete connections	- Option 1: Select the connection and press Delete. - Option 2: Hold Shift and drag with the left mouse button to remove the connection.
Move	Drag on an empty area of the canvas with two fingers, or hold the right mouse button.
Zoom in/out	Use two-finger gestures on the trackpad, or scroll the mouse wheel up and down. Canvas drag shortcut: Use the shortcut in the bottom-right corner of the canvas to switch the left mouse button between Select and Pan Canvas. > Note: On Mac, behavior may vary slightly depending on trackpad and system settings. > >
Run the workflow	Click the Generate button at the bottom to run the entire workflow on the canvas. - Run a single node: Select the node you want to run, then click the button above it to execute that specific node. - Run part of the workflow: Use the left mouse button or a three-finger swipe to select the nodes you want, then click the button above to run only that part of the workflow.
Save generated outputs	- Option 1: Select a node with output and click Download above it to save the result. - Option 2: Open the image/video details, then right-click → Save As to save it to a specific folder.
Save workflows to the cloud	Once a workflow has been successfully run in the workspace, it can be persisted through Cloud Save, so it can be reopened and run again anytime on the platform with greater stability.
View workflow history	Return to the workspace homepage and click My Creations to view all historical workflows.

Use Cases

Video and Storyboarding

Create short-form videos, storyboards, product demos, and branded video content.

With built-in video generation, editing, and templates, Lumina helps teams turn ideas into visual content faster.

E-commerce Content

Generate product images, promotional visuals, and product showcase assets.

For the same product, teams can quickly create multiple versions with different backgrounds, styles, and layouts—ideal for fast-moving e-commerce workflows.

IP and Style-Based Content

Create character-driven content, consistent visual systems, and reusable asset libraries.

When teams need to produce content around fixed characters, defined styles, or brand guidelines, Lumina makes it easier to maintain consistency and scale production over time.

Marketing Content

Create campaign visuals, ad creatives, social content, and promotional assets.

When teams need multiple versions for the same campaign, Lumina helps them move faster while reducing revisions and rework.

Official Components Overview

Basic module

Component	Preview	Description
Basic module	Text generation	The Text node is used to create and manage text content. It supports manual input, AI-generated text, and file import, and is commonly used for prompts, scripts, descriptions, and storyboards. It’s often the starting point of a workflow, with built-in support for editing, previewing, and combining content. Available inputs may vary depending on the model being used.
	Image generation	The Image node is used to generate and display images. It supports text-to-image, image-to-image, image upload, and prompt generation from image. It’s ideal for visual ideation, style exploration, and character or product design. Image outputs can be directly reused across workflows as standard visual assets. Supported inputs may vary depending on the model.
	Video generation	The Video node is used to generate and present dynamic visual content. It supports text-to-video, image-to-video, first-and-last-frame video generation, and video upload. It is suitable for rapid prototyping, scene testing, and motion content creation. It also supports playback and content replacement, making it the core output unit for time-based content. Supported inputs may vary depending on the model service.
	Audio generation	The Audio node is used to generate or process audio content. It can be used for voice, sound effects, or other audio-driven workflows, depending on the model capabilities.
	Multimodal	The Multimodal node allows you to work with text, images, and audio together in one place. By combining instructions with visual inputs, it simplifies tasks such as image editing, style transfer, and text-image creation. It’s well suited for use cases like concept art, product visuals, and creative design workflows. Outputs can be further processed by downstream nodes if needed.
	Agent	The Agent node acts as the brain of the workflow. It understands your instructions, decides what needs to be done, and automatically calls the right tools or workflows to complete the task. It’s especially useful for complex tasks, multi-step workflows, and automation scenarios.
	Composite image	The Image Composition node is used to combine multiple visual elements into a final design. It’s typically used for layouts such as posters, marketing assets, and composed visuals.
	File	The File node is used to upload and preview different types of files. It serves as a flexible input source or reference asset within workflows.
	Recording	The Recording node allows you to capture audio directly within the workflow.
	Multimodal Upload	The Multimodal Upload node is used to upload images, videos, audio, and other media into the workflow.
	Multimodal Preview	The Multimodal Preview node lets you review outputs across different content types before moving forward.
	Font	The Font node allows you to upload and use custom fonts. It outputs a font file path that can be used in nodes like Image Composition for design and text rendering.
Video Generation	Dreamontage_Pro(Multitrack_completion)	This component enables you to combine multiple video or image clips and add transitions between them. It supports effects like long shots, smooth transitions, and AI-generated transitions, making it ideal for more cinematic video creation.
Image Editing	Image SR(AIGC)	Built on the Seed 3.0 model, ImageSR enhances image quality while preserving overall structure. It adds more detail and improves clarity, making it suitable for tasks like image enhancement, creative effects, and old photo restoration.
	Inpainting Edit_V3	This node allows you to edit specific parts of an image by applying a mask. It’s useful for object replacement, background edits, and fine detail adjustments.
Image Generation	DreamO(Single_image)	DreamO helps maintain subject consistency from a single image input. It’s ideal for character consistency, product visuals, and reference-based creation.

Text generation

Module	Item	Description	Preview
Input	Image / Text		Supports multiple image and text inputs. Can be connected via nodes such as Load Image and String to bring in different types of content.
Parameters	Prompt Field		Used to define the generation instruction (prompt). Supports using “@” to reference and insert inputs from other nodes dynamically.
	Model		Supports multiple models, including Doubao series, Qwen, DeepSeek, and others.
Output	Text		Outputs generated text results. Can be connected to nodes like Show Text for display or further processing.

Image generation

Module	Item	Description	Preview
Input	Image / Text		Supports multiple image and text inputs. Can be connected via nodes such as Load Image and String to bring in different types of content.
Parameters	Prompt Field		Used to define the generation instruction (prompt). Supports using “@” to reference and insert inputs from other nodes dynamically.
	Model		Supports the Seedream image generation model series.
Output	Image		Outputs generated images. Can be connected to nodes such as Save Image for storage or further processing.

Video generation

Module	Item	Description	Preview
Input	Image / Text / Audio / Video		Supports multiple inputs, including images, text, audio, and video. Can be connected via nodes such as Load Image and String to bring in different types of content.
Parameters	Prompt Field		Used to define the generation instruction (prompt). Supports using “@” to reference and insert inputs from other nodes dynamically.
	Model		Supports multiple video model series, including Seedance, Wanxiang, and Hailuo.
Output	Video		Outputs generated video content. Can be connected to nodes such as Save Video for storage or further processing.

Audio generation

Supports text-to-speech (TTS) powered by Doubao TTS 2.0.

It generates natural-sounding speech by understanding context—capturing tone, pacing, pauses, and emotion more accurately.

Well suited for:

• AI assistants

• Conversational experiences

• Customer support

• Video voiceovers

Languages supported: Chinese, English, Japanese, Indonesian, and Mexican Spanish.

Note: Make sure the input language matches the selected voice. Mismatches may cause generation to fail.

Parameter	Function	Description	Preview
voice_id	Voice selection		Select the voice for synthesis. Default voice is “James.” Supports Chinese and English, with the option to switch language direction.
speed_ratio	Playback speed		Controls speech speed. 0 = default speed. Positive values make it faster, negative values slower. Useful for matching video pacing or narration style.
loudness_ratio	Volume control		Adjusts overall volume. 0 = default level. Positive values increase volume, negative values decrease it. Suitable for different playback scenarios.
explicit_language	Content filtering		Automatically detects and filters sensitive or non-compliant content in the text to ensure safe output.
model	Model selection		Selects the TTS model. Seed-TTS 1.1 is a high-fidelity model that supports emotion control and cross-language generation.
format	Output format		Defines the export format of the audio file. MP3 is the default and most compatible option. Lossless formats such as WAV are also supported.
sample_rate	Sample rate		Audio sampling rate (Hz). 24,000 Hz is the standard for TTS, balancing audio quality and file size.
bit_rate	Bit rate		Audio bit rate (bps). 32,000 bps (32 kbps) is a commonly used high-quality setting for MP3.
emotion_scale	Emotion intensity		Controls how expressive the voice sounds. Higher values produce stronger emotional expression. Supports fine-grained adjustment for different scenarios.

Multimodal

Module	Item	Description	Preview
Input	Image / Text		Supports multiple image and text inputs. Can be connected via nodes such as Load Image and String to bring in different types of content.
Parameters	Prompt Field		Used to define the generation instruction (prompt). Supports using “@” to reference and insert inputs from other nodes dynamically.
	Model		Supports models such as the Seedream series and Nano Banana.
Output	Image \& Text		Outputs both images and text. Results can be routed to nodes like Show Text and Save Image for display or further processing.

Agent

Item	Description	Preview
Main Model	Acts as the “brain” of the Agent, responsible for reasoning, decision-making, and orchestration within the workflow. - Supports models such as Doubao 1.6, Doubao 1.8, with continuous updates. - Some models support image understanding (indicated by the icon). - Available parameters may vary by model, including: - Thinking: Enables deeper reasoning - Temperature: Controls randomness (higher = more creative, lower = more stable) - Top P: Controls token selection range (lower = more stable output)
Max Iterations	Defines the maximum number of steps the Agent can take internally. Controls how many reasoning or execution steps are allowed per task.
Expand	Expands the Agent node, allowing you to generate, inspect, and adjust its internal structure and outputs.
Seed	Controls randomness. Default value -1 means fully random generation.
Inputs \& Outputs	Supports text, image, video, and audio. Allows: - Multiple inputs of the same type - Mixed multimodal inputs Example: one Agent node can take 4 text inputs + 3 images + 2 videos at the same time.
System Instructions	Defines the Agent’s behavior and role—how it understands tasks, reasons, and produces outputs. Supports using “@” to reference inputs at specific positions for more precise control. Default Prompt: SQL<br>**Role Setup**<br>**Name:** ByteArtiest Assistant <br>**Role:**<br> A general-purpose problem solver.<br> Responsible for understanding user requests, breaking them down, and turning them into clear, actionable steps using available tools—ensuring each step is well-defined, complete, and executable to achieve the intended outcome. <br>**Voice & Tone**<br> Clear, structured, and professional <br> Simple and easy to understand (avoid unnecessary jargon) <br> Adapts tone based on the task: <br> More formal for work-related tasks <br> More friendly for everyday or personal tasks <br> Can be engaging and slightly playful when appropriate <br>**Expertise**<br>1. Task breakdown and workflow design(e.g. work projects, study plans, daily tasks)<br>2. Tool selection and step planning(matching the right tools to the task) <br>3. Logic validation and optimization(ensuring steps are complete, efficient, and executable) <br>4. Cross-domain understanding(turning vague requests into clear, actionable goals) <br>**Goals & Workflow**<br> Always prioritize completing the user’s request effectively and thoroughly. <br>
Task Prompt	The instruction for a specific task (i.e., the prompt). Supports using “@” to reference inputs or tools dynamically, enabling more flexible and controllable execution. > Note: Both system instructions and task prompts guide how the model generates outputs. If no general rules are defined, task-specific prompts can also be placed in system instructions. > >
Tools (Components)	Defines what tools the Agent can use. If connected to X tools, the Agent can use all of them during execution. Without connections, the Agent cannot access those capabilities. In Canvas, tools are essentially components. To let the Agent use a workflow, the workflow must be packaged as a component and connected to the Agent. Tool Description (Optional) Helps the Agent better understand how to use each tool and its parameters. Should include: - Parameter meaning - Whether values are fixed or ranged - Whether fields are required Can be iteratively refined during testing to improve Agent performance. Note: Green indicates successful execution. Gray / Yellow indicates unsuccessful execution.
Output	Results are generated step by step. The number of steps can be controlled via Max Iterations. The Agent breaks down tasks based on the prompt and executes them sequentially to produce the final output.

Composite image

Supports multi-layer composition and editing, with controls for both images and text effects.

Well suited for use cases such as:

• Ad creatives

• Poster design

• Marketing visuals

It also supports preset layouts and automated batch output, making large-scale content production more efficient.

In the full-screen composition view, frequently used layer settings can be pinned and exposed directly on the node for quicker access.

Module	Item	Description	Preview
Input	Image / Font		Supports uploading images and local fonts.
Canvas Settings	Fill Color		Select the fill color for the canvas.
Image Settings	X、Y		Adjust the position of the image on the canvas (left/right, up/down).
	width/height		Resize the image.
	rotation		Rotate the image.
	Flip horizontally		Mirror the image left to right.
	Flip vertically		Flip the image upside down.
	Canvas horizontal align		Align the image horizontally on the canvas (left / center / right).
	Canvas vertical align		Align the image vertically on the canvas (top / center / bottom).
	Flip models		Adjust how the image fits within its bounds (may affect overall layout).
	Rounded corners		Add rounded corners to the image.
	opacity		Adjust transparency (lower = more transparent).
	Blend models		Control how the image blends with layers beneath it (affects color and lighting).
Text Settings	X、Y		Adjust the position of the text on the canvas.
	width/height		Resize the text box (affects wrapping and layout).
	rotation		Rotate the text.
	Flip horizontally		Mirror the text left to right.
	Flip vertically		Flip the text upside down.
	Canvas horizontal align		Align the text box horizontally on the canvas.
	Canvas vertical align		Align the text box vertically on the canvas.
	Text horizontal align		Align text within the text box (left / center / right).
	Text vertical align		Align text vertically within the text box (top / middle / bottom).
	Italic		Apply italic style.
	Underline		Add underline to text.
	deleteline		Add a strike-through effect to text.
	Text control		Controls overall text behavior (e.g. visibility, edit state).
	Font		Select the font family.
	Font size		Adjust text size.
	Line spacing		Adjust spacing between lines.
	Letter spacing		Adjust spacing between characters.
	Fill color		Set text color.
	Blend models		Control how text blends with the background (affects color and visual result).

Font

The Font Node is primarily used together with the Image Composition Node to control text styling in generated visuals.

• Supports selecting fonts from a built-in library or uploading custom fonts

• Enables automated font application within workflows

• Includes 8 preset font styles

• Supports TTF / OTF formats

Module	Item	Description	Preview
Font	Font		Supports uploading custom fonts or selecting from the built-in font library. Can be used with the Image Composition Node to control and adjust text styling in visual outputs. Once connected, the selected font is applied within the workflow, overriding canvas-level font settings.

Recording

Supports recording audio and using it directly within workflows, enabling a complete flow from recording → audio/video composition → export or voiceover.

Typical Use Cases:

• Short-form video creation: Add custom voiceovers to landscape or product videos, and generate results in one step.

• Voice cloning: Record high-quality voice samples to create a personalized AI voice.

• Video dubbing / voiceover: Use recorded audio as a reference to generate voiceovers in a similar style, then merge them into videos.

• Scalable content production: Batch upload videos and batch record or generate audio to quickly produce multiple content variations.

Video Generation

Seedance_1.0mini

Module	Item	Description	Preview
Input	Prompt		Text prompt used to generate the video. Currently supports Chinese only. Recommended to keep within \~500 characters. Longer inputs may be truncated due to model limits.
Parameters	frames		Total number of frames to generate. Default is 121. Must follow the rule: 24 × n + 1 (n = seconds). Examples: - 121 → \~5 seconds - 73 → \~3 seconds Supports videos from 3 to 12 seconds (73–289 frames).
	seed		Random seed. Default is -1. - -1 = fully random - Any other value = fixed seed for reproducible results
	resolution		Output resolution. Default is 480p. Supported options: 480p, 720p, 1080p
	camera_fixed		Keeps the camera static. Default is off (false).
	aspect_ratio		Video aspect ratio. Default is 16:9. Supported options: 16:9, 4:3, 1:1, 3:4, 9:16, 21:9 Different ratios correspond to different output dimensions. ```YAML ' 480p':

Seedance 1.0

Module	Item	Description	Preview
Input	Upload Image		Supports image input. Can be connected via the Load Image node.
Parameters	Prompt		Text prompt used to guide video generation. Currently supports Chinese only.
	Seed		Random seed. - -1 means random generation. - Any other value uses a fixed seed. Range: 0 to 2³¹−1.
	Resolution		Output resolution. Supported options: 480p, 720p, 1080p.
	Camera Fixed		Keeps the camera static. Can be turned on or off.
	Frames		Total number of frames to generate. Must follow 24 × n + 1 (where n = number of seconds). For example: - 121 ≈ 5 seconds - 73 ≈ 3 seconds. Supports 3 to 12 seconds (73–289 frames).
Output	Save Video		Outputs and saves the generated video.

Seedance 2.0 pro fast (Coming soon)

Module	Item	Description	Preview
Input	Upload Image		Supports image input. Can be connected via the Load Image node.
Parameters	Prompt		Text prompt used to guide video generation. Currently supports Chinese only.
	Seed		Random seed. -1 = random. Otherwise uses a fixed value. Range: 0 to 2³¹−1.
	Resolution		Output resolution. Supported: 480p, 720p, 1080p.
	Camera Fixed		Keeps the camera static. Toggle on/off.
	Generate Audio		Whether to include a voice track in the generated video.
	Watermark		Whether to include a watermark in the output video.
	Return Last Frame		Returns the final frame as an image when enabled. Available via the task query API. Output format is PNG, matching the video resolution, with no watermark.
	Tools.Type		Specifies which tools the model is allowed to use during generation.
	Execution Expires After		Task timeout (in seconds). Defines how long the task remains valid after submission, counted from the created_at timestamp.

Seedance 2.0pro (Coming soon)

Module	Item	Description	Preview
Input	Upload Image		Supports image input. Can be connected via the Load Image node.
Parameters	Prompt		Text prompt used to guide video generation. Currently supports Chinese only.
	Seed		Random seed. -1 = random. Otherwise uses a fixed value. Range: 0 to 2³¹−1.
	Resolution		Output resolution. Supported: 480p, 720p, 1080p.
	Camera Fixed		Keeps the camera static. Toggle on/off.
	Generate Audio		Whether to include a voice track in the generated video.
	Watermark		Whether to include a watermark in the output video.
	Return Last Frame		Returns the final frame as an image when enabled. Available via the task query API. Output format is PNG, matching the video resolution, with no watermark.
	Tools.Type		Specifies which tools the model is allowed to use during generation.
	Execution Expires After		Task timeout (in seconds). Defines how long the task remains valid after submission, counted from the created_at timestamp.

Seedance 1.0 tile t2v

Module	Item	Description	Preview
Parameters	Aspect Ratio		Selects the video aspect ratio.
	Frames		Total number of frames. - Default is 121. Must follow 24 × n + 1 (n = seconds). > Examples: 121 ≈ 5s, 73 ≈ 3s. > > - Supports 3–12 seconds (73–289 frames).
	Seed		Random seed. -1 = random. Otherwise uses a fixed value. Range: 0 to 2³¹−1.
	Camera Fixed		Keeps the camera static. Toggle on/off.
	Auto Caption		Automatically expands or rewrites the prompt for better generation results.
	Generate Audio		Whether to include a voice track in the generated video.

Seedance 1.5pro

Module	Item	Description	Preview
Parameters	Return Last Frame		Returns the final frame of the generated video as an image.
	Aspect Ratio		Selects the video aspect ratio.
	Frames per Second		Sets the frame rate (FPS) of the output video.
	Seed		Random seed. -1 = random. Otherwise uses a fixed value. Range: 0 to 2³¹−1.
	Camera Fixed		Keeps the camera static. Toggle on/off.
	Watermark		Whether to include a watermark in the generated video.
	Generate Audio		Whether to include an audio track in the generated video.

Dreamontage pro

Module	Item	Description	Preview
Input	Image		Supports image and video inputs. > Note: set the data type to Binary when uploading media files. > >
Parameters	Prompt		Generates a prompt based on the input media (recommended within 150 characters). If not provided, a prompt will be generated automatically. Recommended within 400 characters. Due to model limits, overly long prompts may be shortened automatically, which can affect the final result.
	Frames		Total number of frames in the generated video.
	Resolution		Currently supports 480p and 720p. The final aspect ratio is automatically selected from 16:9, 4:3, 1:1, 3:4, 9:16, 21:9, and cannot be manually set.
Output	Video		Outputs the generated video. Can be connected to Preview Video and Save Video nodes.

Image Editing

Image SR (AIGC)

Module	Item	Description	Preview
Input	Image		Supports image input. Can be connected via the Load Image node.
Parameters	Prompt		A single prompt used to guide image generation. If left empty, the system will automatically generate a prompt based on the input image.
	Resolution		Supports 2K and 4K output options.
	Final Linear Step		Controls the final stage of the generation process. Range: 0–10.
	Face Protect		Enables face consistency protection. Default: Off.
	Steps		Number of inference steps. Default: 25; Range: 1–100.
	Strength		Controls how strongly the output follows the input image (useful for image-to-image tasks). Range: 0–1.
Output	Images		Outputs generated images and supports saving results.

Inpainting Edit_V3

To create a mask directly on the canvas, right-click the image node and open the Mask Edit tool, then manually select the area to be edited.

Module	Item	Description	Preview
Input	Image		Supports single image input. Can be connected via the Load Image node.
	Mask		Define the area to be edited using a mask. You can upload a mask image directly (black mask), or create one using the Mask Editor within the Load Image node. > Note: The mask area must align with the region you want to edit. > >
Parameters	Prompt		Specify what changes to apply to the selected area.
Output	Image		Outputs the edited image. Can be connected to Preview Image and Save Image nodes.

Image Generation

Seedream5.0

Module	Item	Description	Preview
Parameters	optimize_prompt_options.mode		Controls how prompt optimization is applied. - standard: higher quality results, longer processing time. - fast: faster generation with slightly lower quality.
	tools.type		Defines which tools the model can access and use.
	sequential_image_generation		Controls whether multiple images are generated in sequence. - auto: the model decides whether to generate multiple images and how many, based on the prompt. - disabled: only one image will be generated.
	sequential_image_generation_options.max_images		Sets the maximum number of images to generate per request. Only applies when sequential_image_generation = auto.
	output_format		Sets the output file format (e.g. JPG).
	watermark		Controls whether a watermark is added to the generated images.
	seed		Random seed for generation.
	inner_return_dual_stack_url		Controls whether an internal image URL is returned.

Seedream5.0 lite

Module	Item	Description	Preview
Parameters	force_single		Forces the model to generate a single image only.
	seed		-1 means random generation. Any other value uses a fixed 32-bit positive integer seed.
	min_ratio		Sets the minimum aspect ratio limit. If the automatically selected ratio is lower than this value, min_ratio will be used instead.
	max_ratio		Sets the maximum aspect ratio limit. If the automatically selected ratio is higher than this value, max_ratio will be used instead.
	cot_mode		Controls whether Chain-of-Thought (CoT) mode is enabled. Available options: enable / disable. This setting is not applicable when PE is called manually.
	close_search		Turns off search. By default, search may be used in text-to-image generation.
	use_pre_llm		Enables pre-LLM processing. When enabled, the model can infer resolution more intelligently without requiring a fixed width and height.
	allow_llm_fallback		Allows fallback when PE expansion fails. If enabled, the system will fall back to the original prompt and issue an alert if the fallback affects the result.

Seedream4.5

Module	Item	Description	Preview
Parameters	negative_prompt		Describes what you want to avoid in the generated image.
	seed		Random seed. Use -1 for a random result.
	sequential_image_generation		Controls whether the model can generate multiple images in one request. - auto: the model decides whether to return multiple images and how many, based on the prompt. - disabled: only one image will be generated.
	max_images		Sets the maximum number of images that can be generated in one request. Only takes effect when sequential_image_generation is set to auto.
	watermark		Controls whether a watermark is added to the generated image.
	optimize_prompt_options_mode		Controls how prompt optimization is applied. - standard: higher-quality results, but slower. - fast: faster generation, with slightly lower quality.

Seedream3

Module	Item	Description	Preview
Input	image		Supports image input. Can be connected via the load image node.
Parameters	prompt		Text prompt used to generate the image.
	Prompt Expansion		Uses an LLM to automatically expand and refine the input prompt. Default: True
	Expansion Mode		Defines how the prompt expansion is applied. Options: std (default) / poster
	seed		Random seed. Use -1 for random results; other values fix the seed. Range: [-1, 2³¹−1]
	width		Output image width. Default: 1024 Supported resolution ranges from 512×512 to 2048×2048. Maximum aspect ratio: 1:16
	height		Output image height. Default: 1024 Supported resolution ranges from 512×512 to 2048×2048. Maximum aspect ratio: 1:16
	scale		Controls how closely the output follows the prompt. Default: 2.5
	i2i_strength		Controls how much the output deviates from the input image. Lower values preserve more of the original image.
Output	images		Outputs generated images; can be connected to save image nodes.

Seedream4

Module	Item	Description	Preview
Input	image		Supports image input. Can be connected via the load image node.
Parameters	prompt		Text prompt used to generate the image.
	negative_prompt		Negative prompt used to describe what should be avoided in the generated image. Default: “nsfw”
	seed		Random seed. Use -1 for a random result; any other value sets a fixed seed. Default: -1 Values other than -1 are mapped to a positive int32 integer.
	width		Output image width. Default: 1024 Width × Height must be within [512×512, 2048×2048] • Maximum aspect ratio: 1:16
	height		Output image height. Default: 1024 Width × Height must be within [512×512, 2048×2048] • Maximum aspect ratio: 1:16
	Force Single Output		Controls whether the result is forced to output as a single image only.
Output	images		Outputs generated images and supports saving results.