> ## Documentation Index
> Fetch the complete documentation index at: https://docs.krea.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Nodes
Create powerful AI workflows in Krea’s visual builder to chain models, apply edits and effects, automate repeated steps, and turn workflows into shareable apps.
[**Try Nodes**](https://www.krea.ai/nodes)\
[**krea.ai/nodes**](https://www.krea.ai/nodes)
# The basics
Every node has three key elements:
| **Element** | **Description** |
| :------------- | :---------------------------------------------------------------------------------------------------------------------------- |
| **Inputs** | Connection points on the left side where data flows in. These can be connected to other nodes' outputs or filled in manually. |
| **Parameters** | Settings within the node itself, like strength values, resolution options, or prompt text. |
| **Outputs** | Connection points on the right side where results flow out to other nodes. |
Nodes connect through compatible handles, color-coded by data type (image, text, video, etc.). Drag from any handle to see a list of connectable nodes. You can connect up to 10 outgoing connections from a single node.
## The canvas
The nodes canvas is your workspace for building workflows. You can navigate the canvas using the following controls:
| **Action** | **Control** |
| :------------------ | :------------------------------------------------------------------------------------- |
| **Pan** | Click and drag empty space, or hold spacebar and drag |
| **Zoom** | Scroll wheel or pinch gesture |
| **Select multiple** | Click and drag to draw selection box, or hold Shift and click to select multiple nodes |
There are two navigation modes you can switch between in the left sidebar: **Pan** and **Select**. Clicking and dragging in Pan mode will move the canvas, while Select mode will select nodes.
* To select nodes in Pan mode: hold **Cmd/Ctrl** and click & drag
* To pan the canvas in Select mode: hold **Spacebar** and click & drag
## Organizing
* Use **section nodes** to group related operations visually
* Create **node groups** by selecting multiple nodes and pressing the Group button. Once you have a group, you can drag nodes in or out to add or remove them
* Add **sticky notes** to document your workflow logic
# All available nodes
***Last Updated: April 13, 2026***\
\
This section provides a comprehensive reference of all nodes available in the Krea Nodes workflow editor. Models rotate quarterly—always verify current availability in the Nodes interface at [krea.ai/nodes](https://www.krea.ai/nodes).
## Generate Image
AI models for generating images from text prompts and image references.
| Name | Description | Supports Styles | Image Prompt | Max Image Refs |
| --------------------- | ---------------------------------------------------------------------------- | --------------- | ------------ | -------------- |
| **Krea 1** | Fast creative model, best for aesthetic images and photorealism | Yes | Yes | 2 |
| **Nano Banana Pro** | Newer model with native 4K image generation and editing capabilities | No | Yes | 15 |
| **Nano Banana** | Smart model, good for image editing | No | Yes | 3 |
| **ChatGPT Image 1.5** | Highest quality with best prompt adherence; ideal for logos, icons, and text | No | Yes | 15 |
| **ChatGPT Image** | Highest quality with best prompt adherence | Yes | Yes | 15 |
| **Flux 2** | FLUX.2 \[dev] with enhanced realism and native editing capabilities | No | Yes | 3 |
| **Flux 2 Pro** | BFL's next generation model with improved quality and prompt adherence | No | Yes | 10 |
| **Flux 2 Flex** | BFL's next generation model, excelling at rendering text and fine details | No | Yes | 10 |
| **Flux 2 Max** | BFL's next generation model, excelling at rendering text and fine details | No | Yes | 10 |
| **Flux** | Fastest and cheapest model | Yes | Yes | 2 |
| Flux 1.1 Pro | Advanced yet efficient model from BFL | No | No | - |
| Flux 1.1 Pro Ultra | BFL's highest quality text-to-image model | No | No | - |
| **Z Image** | Fast high quality image model from Alibaba | Yes | Yes | 1 |
| **Qwen Image 2512** | Enhanced realism, finer natural detail, and improved text rendering | Yes | No | - |
| **Seedream 4.5** | Latest high quality model for photorealism and text rendering | No | Yes | 15 |
| **Seedream 4** | New high quality model for photorealism and text rendering | No | Yes | 6 |
| Seedream 3 | New fast, high-quality model from ByteDance | No | No | - |
| **Wan 2.2** | Slow model with great ultra-realistic textures for aesthetic outputs | Yes | No | - |
| **Qwen** | Great text rendering and prompt adherence | Yes | No | - |
| **Flux.1 Krea** | Distilled and open sourced version of Krea 1 | Yes | Yes | - |
| **Flux Kontext** | Frontier model designed for image editing, optimized for Krea | No | Yes | 1 |
| **Flux Kontext Pro** | Frontier model designed for image editing | No | Yes | 4 |
| Imagen 4 Ultra | Google's best image model | No | No | - |
| Imagen 4 Fast | Google's fastest image model | No | No | - |
| Imagen 4 | Google's current generation image model | No | No | - |
| Imagen 3 | Google's previous generation image model | No | No | - |
| Runway Gen-4 | Cinematic image model with references | No | No | - |
| **Kling O1** | High quality image model with reference support | No | No | 10 |
| **Ideogram 3.0** | Highly aesthetic, general-purpose model | No | Yes | 1 |
## Generate Video
AI models for generating videos from text prompts and images.
| Name | Description | Start Image | End Image | Camera Control | Audio | Durations |
| -------------------- | -------------------------------------------------------------------------------------------------- | ----------- | --------- | -------------- | ----- | ------------ |
| **Krea Realtime** | New real-time video generation trained by Krea. Fast and long generations, up to 15 seconds | No | No | No | No | 5s default |
| **Hailuo 2.3** | Frontier model with dynamic motion | Yes | No | No | - | 6s, 10s |
| **Hailuo 2.3 Fast** | Cheapest medium-quality model. Best for most use cases | Yes | No | No | - | 6s, 10s |
| **Hailuo 02** | Frontier model with dynamic motion | Yes | Yes | No | - | 6s, 10s |
| **Hailuo** | High-quality model with camera control | Yes | No | Yes | - | - |
| **Veo 3.1** | Highest-quality frontier model with audio and reference images | Yes | Yes | No | Yes | 4s, 6s, 8s |
| **Veo 3.1 Fast** | Faster frontier model with audio | Yes | Yes | No | Yes | 4s, 6s, 8s |
| **Veo 3** | Highest-quality frontier model with audio | Yes | No | No | Yes | 4s, 6s, 8s |
| **Veo 3 Fast** | Fast frontier model with audio | Yes | No | No | Yes | 4s, 6s, 8s |
| Veo 2 | Expensive high-quality model from Google | Yes | No | No | - | 5-8s |
| **Sora 2** | OpenAI's new powerful video model. Create rich detailed, dynamic clips | Yes | No | No | No | 4s, 8s, 12s |
| **Sora 2 Pro** | OpenAI's state-of-the-art, most advanced video generation model | Yes | No | No | No | 4s, 8s, 12s |
| Wan 2.1 | Fastest lower-quality model with custom styles | Yes | Yes | No | Yes | - |
| **Wan 2.2** | Fast, high-quality model from Alibaba | Yes | No | No | No | - |
| **Wan 2.5** | Latest frontier model from Alibaba with improved quality | Yes | No | No | Yes | 5s, 10s |
| **Wan 2.6** | Latest frontier model from Alibaba with improved quality and multi-shot support | Yes | No | No | Yes | 5s, 10s, 15s |
| **Seedance 1.5 Pro** | Latest frontier model from ByteDance with audio support and end frame | Yes | Yes | No | Yes | 4-12s |
| Seedance Pro Fast | Fast and cheap model. Up to 12 seconds | Yes | No | No | - | 2-12s |
| Seedance Lite | Fast, high-quality model from ByteDance | Yes | Yes | No | - | 5s, 10s |
| Seedance Pro | Fast, high-quality model from ByteDance | Yes | Yes | No | - | 5s, 10s |
| **Kling o1** | Reasoning video model that thinks before generating. Supports image, element, and video references | Yes | Yes | No | - | 3-10s |
| **Kling 2.6** | Newest frontier model from Kling with native audio | Yes | Yes | No | Yes | 5s, 10s |
| Kling 2.5 | Next-gen model with improved dynamics and enhanced style adaptation | Yes | Yes | No | - | 5s, 10s |
| Kling 2.5 Turbo | Top-tier text-to-video with unparalleled motion fluidity | Yes | Yes | No | - | 5s, 10s |
| Kling 2.1 | Frontier model with 1080p resolution | Yes | Yes | No | - | 5s, 10s |
| Kling 2.0 | High-quality model with great aesthetics | Yes | No | No | - | 5s |
| Kling 1.6 | High-quality model for complex scenes | Yes | No | No | - | 5s, 10s |
| Kling 1.0 (Pro) | Slow model with high control and 10s duration | Yes | Yes | No | - | 5s, 10s |
| **LTX-2** | High-quality audio-video model from Lightricks with native sound generation | Yes | No | No | No | 5s, 8s, 10s |
| **Vidu Q2** | High-quality model with reference images support | Yes | No | No | - | 2-8s |
| **01-Live** | High-quality model for animating people | Yes | No | No | - | - |
| **Runway Gen-4** | High-quality, cinematic model | Yes | No | No | - | 5s, 10s |
| **Runway Gen-3** | Cinematic model with high consistency | Yes | No | No | - | 5s, 10s |
| Hunyuan | Fast, inexpensive model with live previews | No | No | No | - | - |
| Ray 2 | Fast next-gen model with natural motion | Yes | No | No | - | - |
## Edit Image
Models for editing and modifying existing images.
| Name | Description | Inpainting | Outpainting | Prompt Editing | Reference Images |
| ---------------------- | ------------------------------------------------------------------------------ | ---------- | ----------- | -------------- | ---------------- |
| **Flux** | More tools, more control. Define exact regions to change, or expand your image | Yes | Yes | No | Yes |
| **Nano Banana Pro** | Google's state-of-the-art image editing model | No | No | Yes | Yes |
| **Nano Banana** | Google's powerful image editing model | No | No | Yes | Yes |
| **ChatGPT Image 1.5** | OpenAI's latest image model with improved quality and prompt adherence | No | No | Yes | Yes |
| **Flux 2 Pro** | BFL's next generation model with improved quality and prompt adherence | No | No | Yes | Yes |
| **Flux 2 Flex** | BFL's next generation model, excelling at rendering text and fine details | No | No | Yes | Yes |
| **Flux 2 Max** | BFL's next generation model with maximum performance | No | No | Yes | Yes |
| **Qwen Image Plus** | New large image editing model from Alibaba | No | No | Yes | Yes |
| Qwen Image Edit | Advanced image editing model from Alibaba | No | No | Yes | No |
| **Flux Kontext** | Frontier model designed for image editing | No | No | Yes | No |
| **Flux Kontext Pro** | Fast iterative editing with character consistency | No | No | Yes | No |
| **Flux Kontext Max** | Maximum performance with improved prompt adherence and typography | No | No | Yes | Yes |
| **ChatGPT Image** | OpenAI's most advanced image model | No | No | Yes | Yes |
| **SeedEdit** | New frontier model designed for image editing (character references) | No | No | Yes | Character Refs |
| **Ideogram** | New frontier model designed for inpainting | Yes | No | No | Character Refs |
| **Qwen Image Layered** | Decompose your image into semantic layers | No | No | Yes | No |
## Enhance Image
Models for upscaling and enhancing image quality.
| Name | Description | Max Resolution |
| --------------------- | -------------------------------------------------------------------------------------- | -------------- |
| **Krea Enhance** | Krea's powerful generative upscaler. Best for adding new details & improving AI images | 8192px |
| **Upscale V1** | Simple, fast upscaler that preserves details exceptionally well | 8000px |
| **Bloom** | Creatively upscale images up to 8x, while adding stunning detail | 10000px |
| **Magnific Creative** | Magnific AI upscaler with presets for portraits, art, landscapes | 10000px |
| **Magnific Precise** | Magnific Precise V2 upscaler with sharpening, grain, and ultra detail controls | 10000px |
| **Topaz** | Powerful upscaler with advanced settings. Up to 22K resolution | 22000px |
| **Topaz Generative** | Slower high quality enhancer with powerful settings | 16000px |
| Krea Legacy | Krea's generative legacy enhancer with scene transfer | 4096px |
## Enhance Video
Models for upscaling and enhancing video quality.
| Name | Description | Max Resolution |
| --------------- | --------------------------------------------------------------------------------------------- | -------------- |
| **Topaz Video** | Topaz's powerful AI upscaler. Add frames and grain, fix focus, and more | 8000px |
| **Starlight** | Diffusion-based AI upscaler that transforms low-resolution and degraded video into HD quality | 4000px |
| **Astra** | Creative video enhancer for upscaling AI-generated videos up to 4K resolution | 4000px |
| **SeedVR2** | ByteDance's fast video upscaler for quick video enhancement | 3840px |
| **Krea Video** | Krea's generative upscaler. Best for adding new details & improving AI videos | 4096px |
## Generate 3D
AI models for generating 3D objects.
| Name | Description |
| --------------------- | ------------------------------------------------------------- |
| **TRELLIS 2** | Native 3D generative model for high-quality 3D asset creation |
| TRELLIS | High quality 3D model generator |
| **Tripo 3.0** | New state-of-the-art 3D model generator by Tripo |
| Tripo | High quality 3D model generator |
| Hunyuan3D-2mini-Turbo | Fastest 3D model generator |
| Hunyuan3D-2 | High quality 3D model generator |
| Hunyuan3D-2.1 | High quality 3D model generator |
## Motion Transfer
Transfer motion from one video to another.
| Name | Description |
| --------------------------- | --------------------------------------------------------------------- |
| **Kling Motion Control** | Transfer motion from one video to another character |
| **Runway Aleph** | Runway's Gen-4 Aleph model for advanced video-to-video transformation |
| **Wan 2.2 Animate Replace** | Replace characters in videos while preserving scene context |
| **Wan 2.2 Animate Move** | Transfer motion from one video to another character |
## Lipsync
Synchronize lip movements with audio.
| Name | Description |
| ---------- | -------------------------------------------- |
| **Fabric** | Turn any image into a talking video. By VEED |
| **Hedra** | Omnimodal lipsync model by Hedra |
## Audio
Nodes for audio generation and synthesis.
| Node Type | Description |
| -------------- | ---------------------------------------------------------------------------- |
| **Elevenlabs** | Generate audio with Elevenlabs text-to-speech. Write text and select a voice |
| **ThinkSound** | Generate synchronized sound, speech and soundscapes for a video |
| **MMAudio** | Transform silent videos into immersive experiences with audio-synthesis |
## Utility
Utility nodes that help you process images, videos, text, and more. Most utility nodes require a Pro subscription.
### Text Utility
| Node Type | Description |
| ----------------- | --------------------------------------------------------------------------------- |
| **LLM Call** | Generate text with a LLM. Supports GPT-5, GPT-5-Mini, and GPT-4o-Mini, etc. |
| **Line Splitter** | Split text into lines |
| **Concat Text** | Concatenate multiple text inputs into one. Supports multiple join operators |
| **Sticky Note** | Add a sticky note to the canvas |
| **Text Overlay** | Add text overlay on an image with customizable font, size, color, and positioning |
### Image Utility
| Node Type | Description |
| ------------------------- | ---------------------------------------------------------------- |
| **Blur Image** | Blur the image |
| **Invert Image** | Invert the colors of an image |
| **Brightness & Contrast** | Adjust both brightness and contrast of an image in a single node |
| **Compositor** | Blend two image layers together with adjustable opacity |
| **Hue & Saturation** | Adjust the hue and saturation of an image |
| **RGB Adjust** | Independently adjust the red, green, and blue color channels |
| **Remove Background** | Remove the background from an image |
| **Crop Image** | Extract a specific region from an image by cropping |
| **Image Mask Editor** | Edit the mask of an image |
### Video Utility
| Node Type | Description |
| -------------------------- | ------------------------------------------------------------------------------- |
| **Get Video Frame** | Extract a frame from a video. Supports the first or last frame |
| **Trim Video** | Trim a video to a specific time range |
| **Trim Audio** | Trim audio to a specific time range |
| **Video Speed** | Adjust video playback speed |
| **Combine Video & Audio** | Combine a video and audio file into a single video |
| **Video Time Ramp** | Remap video timing with different curve types (ease-in, ease-out, bezier, etc.) |
| **Stitch Videos** | Stitch multiple videos together. Supports multiple video inputs |
| **Video Hue & Saturation** | Adjust the hue and saturation of a video |
| **Crop Video** | Crop a video to a specific width and height |
| **Qwen Edit Camera** | Edit an image using the Qwen Edit Camera model |
> **Note:** Models are ordered by popularity and release date. Bolded entries represent the most actively used models. Availability and features may change as Krea adds new models quarterly.
>
> **API Access:** All models available via API at [docs.krea.ai/api-reference](https://docs.krea.ai/api-reference)
# Node
# Utility nodes
Utility nodes provide general-purpose functionality for manipulating data, organizing workflows, and preparing content for AI models. They form the “supporting infrastructure” of your pipeline, handling common tasks without running AI inference themselves.
## What utility nodes are for:
* **Do not run AI models** — they don’t perform AI inference or external API calls directly
* **Provide essential support functions**, including:
* State manipulation (transform, update, clean variables)
* Workflow composition (organize, group, reroute connections)
* Data formatting (text, images, video preprocessing)
* Math and logic operations
* Graph management (navigation, display)
They let you manipulate state variables, run utility transformations, and compose workflows from reusable components.
## Categories of utility nodes
Krea utility nodes are organized into three main categories:
### 1. Text utility nodes
For manipulating text inputs and outputs:
| **Node** | **Purpose** |
| :---------------- | :--------------------------------------------------------------------------------------- |
| **LLM Call** | Generate text with a large language model. Supports GPT-5, GPT-5-Mini, GPT-4o-Mini, etc. |
| **Line Splitter** | Split text into individual lines based on delimiters |
| **Concat Text** | Concatenate multiple text inputs into one; supports multiple join operators |
| **Sticky Note** | Add a visual note to the canvas for documentation |
| **Text Overlay** | Add text on images with customizable font, size, color, and positioning |
**Common use cases:**
* Split prompts into batches for sequential processing
* Combine metadata fields into a single formatted string
* Add captions or watermarks to generated images
### 2. Image utility nodes
For post-processing and preparing image inputs:
| **Node** | **Purpose** |
| :------------------------ | :------------------------------------------------------- |
| **Blur Image** | Apply Gaussian blur to an image |
| **Invert Image** | Invert all colors for negative effects |
| **Brightness & Contrast** | Adjust both brightness and contrast in a single node |
| **Compositor** | Blend two image layers with adjustable opacity |
| **Hue & Saturation** | Adjust hue and saturation of an image |
| **RGB Adjust** | Independently adjust red, green, and blue color channels |
| **Remove Background** | AI-powered background removal for product photography |
| **Crop Image** | Extract a specific region by cropping |
| **Image Mask Editor** | Edit the mask of an image (for inpainting/segmentation) |
**Common use cases:**
* Color correction between generation and upscaling
* Background removal for product compositing
* Adjusting hues/saturation for brand consistency
* Creating watermarked overlays with compositor node
## 3. Video utility nodes
| **Node** | **Purpose** |
| :------------------------- | :-------------------------------------------------------- |
| **Get Video Frame** | Extract first or last frame from a video |
| **Trim Video** | Trim video to a specific time range |
| **Trim Audio** | Trim audio to a specific time range |
| **Video Speed** | Adjust playback speed (slow/fast) |
| **Combine Video & Audio** | Combine video and audio into a single file |
| **Video Time Ramp** | Remap timing with curve types (ease-in, ease-out, bezier) |
| **Stitch Videos** | Stitch multiple videos together; supports multiple inputs |
| **Video Hue & Saturation** | Adjust hue and saturation of video |
| **Crop Video** | Crop to specific width and height |
| **Qwen Edit Camera** | Edit an image using the Qwen Edit Camera model |
**Common use cases:**
* Extracting frames from generated videos for thumbnails
* Trimming clips to specific lengths for social media
* Combining AI-generated video with Elevenlabs audio
* Speed ramping for cinematic effects
## Best Practices for Utility Nodes
1. **Organize with groups** — Select multiple nodes and group them to create logical sections
2. **Label with sticky notes** — Document workflow logic for yourself or team members
3. **Use state modifiers early** — Clean and prepare data before it reaches AI models
4. **Keep single responsibility** — Each utility node should do one thing well; chain them for complex workflows
5. **Preview at each step** — Use display nodes to check intermediate outputs (e.g., Display Any from rgthree)
# Node app builder
Krea's Node App Builder lets you turn complex workflows into simple, **user-friendly apps with custom input forms and clean interface**s, no coding required.
Think of it this way: you spend hours building a sophisticated workflow with 30+ nodes (image generation → style transfer → background removal → upscaling → watermark → video composition). Instead of sharing that intimidating graph with your client or team, you package it as an app where they just:
1. **Upload an image** (or enter a prompt)
2. **Pick 2-3 settings** from dropdowns
3. **Click "Generate"**
4. **Get the result**
Your team never see the 47 nodes, the connections, or know which models are running. Just inputs → magic → output.
## How it works
### Step 1: Build your workflow
Create your full pipeline in the Nodes editor. Wire up everything you need: multiple models, utility nodes, loops, conditionals, whatever your process requires.
### **Step 2: Define inputs and outputs**
Mark which node parameters should be exposed as user inputs:
* Text prompts
* Image upload slots
* Dropdown selects (style options, models, aspect ratios)
* Sliders (strength, quality, duration)
* Number inputs (quantity, resolution)
### Step 3: Design the app interface
Customize the simple UI that users will see:
* Field labels and descriptions
* Input order
* Default values
* Required vs. optional fields
### Step 4: Publish and share
Get a shareable link or embed code:
* **Public apps**: Anyone with the link can use them
* **Private apps**: Share only with team members
* **API access**: Trigger via the Execute a Node App endpoint for custom integrations
## Real-World Use Cases
### Agencies → Clients
* **Product photo generator**: Agency builds a 50-node workflow for e-commerce photo shoots. Client uploads a product photo, picks a background style (beach, studio, minimalist), and gets 10 polished images in 30 seconds.
* **Social media content pack**: Client enters a product name → app generates 12 social posts with images, captions, and hashtags pre-formatted for Instagram, TikTok, and LinkedIn.
### Teams → Internal Users
* **Brand-compliant asset generator**: Marketing team builds a workflow enforcing brand guidelines (colors, fonts, logo placement). Any team member can generate on-brand visuals without design skills.
* **Video repurposing pipeline**: Content team uploads one long video → app automatically creates Shorts/Reels/TikToks with captions, branded intros/outros, and platform-specific aspect ratios.
### Enterprises → Operations
* **Document processing workflow**: Image → OCR → data extraction → formatting → database entry. Non-technical staff just drop in documents.
* **Customer onboarding automation**: New customer info → personalized welcome video → customized product mockups → email template generation.
## Technical Details
**What gets exposed as app inputs:**
* Text inputs (prompts, descriptions)
* Image uploads (with optional preview)
* Number fields (with min/max constraints)
* Dropdowns/selects (predefined options from model parameters)
* Boolean toggles (yes/no, on/off)
* Sliders (mapped to node parameter ranges)
**What stays hidden:**
* Intermediate nodes (utilities, transformations)
* Model switching logic
* Error handling workflows
* Conditional branching
**Pricing:**
* **Pro plan**: Unlocks Nodes + Apps access
* **Max plan** (\$105/month, 60,000 compute units): Unlimited concurrency—run multiple apps/workflows simultaneously
* **Business plan** (up to 50 seats): Share private Node Apps with your team
* Each app execution costs compute units based on the models used (shown before running via Node Agent)
**API integration:**\
The Execute a Node App endpoint lets you trigger apps programmatically, perfect for:
* Building custom frontends
* Integrating into existing tools
* Automating enterprise pipelines
## Best Practices
**Design your app for your users, not yourself:**
* **Limit input fields to 5-7 max**. If you need more, you're asking too much of users.
* **Use descriptive labels**. Instead of "strength" use "How intense should the style transfer be?"
* **Set smart defaults**. Pre-select the most common options so users get results quickly without thinking.
* **Hide power-user options**. Put advanced settings behind an "Advanced" toggle or leave them internal.
* **Test with non-technical people**. If they can't use it in under 2 minutes without asking questions, simplify the interface.
# Krea Node Agent
Agentic AI has mostly lived in developer tools until now: Krea's **Node Agent** turns a sentence into a runnable creative workflow. You type what you want to make. It reads your canvas, plans the pipeline, wires the nodes, and runs the job.
## Starting a Workflow
Write a natural language description of what you want to make, such as:
* *"cinematic video of a futuristic city at sunset"*
* *"generate four variations of a portrait using different styles, then upscale the best one"*
* *"take an input image, remove the background, and composite it onto three different scenes"*
The agent starts by **reading your canvas**. Whatever's already there, it factors in:
* Existing nodes
* Connections
* Outputs from earlier runs
If you have a **style node from a previous session**, it reuses that instead of creating a new one.
Most AI tools treat every request as a blank slate. The Node Agent doesn't. It sees your full canvas state, not just the words you typed.
## Plan, Then Build
### Review the Plan
The agent shows you a **plan before it touches anything**. For a futuristic city video, the plan might be four stages:
1. **Prompt refinement**
2. **Image generation** (on K1 or Flux)
3. **Enhancement**
4. **Video generation**
You can:
* **Swap models** before building
* **Drop stages** you don't need
* **Approve the plan** to start building
### **Watch It Build**
Once you approve, **nodes appear on your canvas one layer at a time**. Each one wires into the last as it's placed. You see the workflow assemble in real time, not just a final result.
## Validation and cost
### Automatic validation
Before the job runs, the agent **walks the full graph** and checks everything:
| **Problem** | **Agent's Fix** |
| :-------------------------------------------------- | :-------------------------------------------------------------- |
| Missing parameter | Fills it in automatically |
| Incompatible connection (output → wrong input type) | Fixes the connection **or** adds a conversion node between them |
### Transparent cost
Then it shows you **compute cost, broken down per node**. Nothing runs until you say go.
> **You see what it will cost before anything processes.**
## Best practices
### Writing effective prompts
| **Do This** | **Instead of This** |
| :--------------------------------------------------------------------------------- | :------------------------------------------------ |
| Be specific: *"cinematic video of a futuristic city at sunset in cyberpunk style"* | Vague: *"make a video"* |
| Include constraints: *"four portrait variations, upscalled to 4K"* | Unspecified: *"make portraits, then fix them"* |
| Reference existing work: *"use the same style node I used yesterday"* | Ignore canvas: *"create everything from scratch"* |
### Leveraging canvas state
The more work you've already put on your canvas, the smarter the agent gets:
* **Style nodes** from previous sessions get reused
* **Existing models** are factored into the plan
* **Cached outputs** prevent redundant processing
* **Grouped workflows** are treated as single units
### Iteration patterns
1. **Start broad**: Let the agent build a complete pipeline from a simple description
2. **Tweak incrementally**: Change one parameter and watch only affected nodes rerun
3. **Banish stages**: Reject expensive or unnecessary stages before building
4. **Swap models**: Try Flux instead of K1 without rebuilding the whole graph
5. **Extend workflows**: Add new stages (e.g., video after image generation) without starting over
### Plan requirements
Node Agent is available with **Nodes access**:
| **Plan** | **Features** |
| :---------------------------- | :------------------------------------------------------------------------------------------------------------- |
| **Pro** (\~\$30/month) | Unlocks Nodes + Apps; Node Agent access |
| **Max** (\$105/month) | 60,000 compute units/month; **unlimited concurrency** for running multiple Node Agent workflows simultaneously |
| **Business** (up to 50 seats) | Share private Node Apps with your team; collaborative canvas editing |
Each model invocation costs compute units. The agent shows cost before building so you can adjust the plan if needed (drop stages, swap cheaper models).
## Troubleshooting
| **Issue** | **Solution** |
| :---------------------------- | :--------------------------------------------------------------------------------------------------------------------------------------- |
| **Agent shows no plan** | Make your prompt more specific. Include content type (video/image), style (cyberpunk/realistic), and constraints (quantity, resolution). |
| **Nodes fail to connect** | The agent should auto-fix this. If not, manually check if the output and input types match (image→image, text→text). |
| **Cost is too high** | Reject expensive stages (e.g., enhancement, video) before building. Swap to cheaper models (Flux vs. Flux 2 Pro). |
| **Reruns the whole workflow** | Check if you edited a node early in the chain. The agent should only rerun downstream nodes. Clear cache if needed. |
| **Missing parameter error** | Click the node to see which parameter is missing.agent should auto-fill most of them. |
## Quick Reference
| **Feature** | **Description** |
| :------------------------- | :---------------------------------------------------------------- |
| **Prompt-to-Workflow** | Type a sentence → get a complete pipeline |
| **Canvas-Aware Planning** | Reads existing nodes, reuses styles, factors in cached outputs |
| **Plan Before Building** | Review and approve stages before anything runs |
| **Auto-Validation** | Fills missing parameters, fixes incompatible connections |
| **Cost Transparency** | See compute cost per node before building |
| **Downstream-Only Reruns** | Only affected nodes reprocess when you edit |
| **Model Selection** | Agent picks K1/Flux/Imagen/video/enhancement models automatically |
| **Real-Time Assembly** | Nodes appear and wire together as the agent builds |
### 2. Image utility nodes
For post-processing and preparing image inputs:
| **Node** | **Purpose** |
| :------------------------ | :------------------------------------------------------- |
| **Blur Image** | Apply Gaussian blur to an image |
| **Invert Image** | Invert all colors for negative effects |
| **Brightness & Contrast** | Adjust both brightness and contrast in a single node |
| **Compositor** | Blend two image layers with adjustable opacity |
| **Hue & Saturation** | Adjust hue and saturation of an image |
| **RGB Adjust** | Independently adjust red, green, and blue color channels |
| **Remove Background** | AI-powered background removal for product photography |
| **Crop Image** | Extract a specific region by cropping |
| **Image Mask Editor** | Edit the mask of an image (for inpainting/segmentation) |
**Common use cases:**
* Color correction between generation and upscaling
* Background removal for product compositing
* Adjusting hues/saturation for brand consistency
* Creating watermarked overlays with compositor node
## 3. Video utility nodes
| **Node** | **Purpose** |
| :------------------------- | :-------------------------------------------------------- |
| **Get Video Frame** | Extract first or last frame from a video |
| **Trim Video** | Trim video to a specific time range |
| **Trim Audio** | Trim audio to a specific time range |
| **Video Speed** | Adjust playback speed (slow/fast) |
| **Combine Video & Audio** | Combine video and audio into a single file |
| **Video Time Ramp** | Remap timing with curve types (ease-in, ease-out, bezier) |
| **Stitch Videos** | Stitch multiple videos together; supports multiple inputs |
| **Video Hue & Saturation** | Adjust hue and saturation of video |
| **Crop Video** | Crop to specific width and height |
| **Qwen Edit Camera** | Edit an image using the Qwen Edit Camera model |
**Common use cases:**
* Extracting frames from generated videos for thumbnails
* Trimming clips to specific lengths for social media
* Combining AI-generated video with Elevenlabs audio
* Speed ramping for cinematic effects
## Best Practices for Utility Nodes
1. **Organize with groups** — Select multiple nodes and group them to create logical sections
2. **Label with sticky notes** — Document workflow logic for yourself or team members
3. **Use state modifiers early** — Clean and prepare data before it reaches AI models
4. **Keep single responsibility** — Each utility node should do one thing well; chain them for complex workflows
5. **Preview at each step** — Use display nodes to check intermediate outputs (e.g., Display Any from rgthree)
# Node app builder
Krea's Node App Builder lets you turn complex workflows into simple, **user-friendly apps with custom input forms and clean interface**s, no coding required.
Think of it this way: you spend hours building a sophisticated workflow with 30+ nodes (image generation → style transfer → background removal → upscaling → watermark → video composition). Instead of sharing that intimidating graph with your client or team, you package it as an app where they just:
1. **Upload an image** (or enter a prompt)
2. **Pick 2-3 settings** from dropdowns
3. **Click "Generate"**
4. **Get the result**
Your team never see the 47 nodes, the connections, or know which models are running. Just inputs → magic → output.
## How it works
### Step 1: Build your workflow
Create your full pipeline in the Nodes editor. Wire up everything you need: multiple models, utility nodes, loops, conditionals, whatever your process requires.
### **Step 2: Define inputs and outputs**
Mark which node parameters should be exposed as user inputs:
* Text prompts
* Image upload slots
* Dropdown selects (style options, models, aspect ratios)
* Sliders (strength, quality, duration)
* Number inputs (quantity, resolution)
### Step 3: Design the app interface
Customize the simple UI that users will see:
* Field labels and descriptions
* Input order
* Default values
* Required vs. optional fields
### Step 4: Publish and share
Get a shareable link or embed code:
* **Public apps**: Anyone with the link can use them
* **Private apps**: Share only with team members
* **API access**: Trigger via the Execute a Node App endpoint for custom integrations
## Real-World Use Cases
### Agencies → Clients
* **Product photo generator**: Agency builds a 50-node workflow for e-commerce photo shoots. Client uploads a product photo, picks a background style (beach, studio, minimalist), and gets 10 polished images in 30 seconds.
* **Social media content pack**: Client enters a product name → app generates 12 social posts with images, captions, and hashtags pre-formatted for Instagram, TikTok, and LinkedIn.
### Teams → Internal Users
* **Brand-compliant asset generator**: Marketing team builds a workflow enforcing brand guidelines (colors, fonts, logo placement). Any team member can generate on-brand visuals without design skills.
* **Video repurposing pipeline**: Content team uploads one long video → app automatically creates Shorts/Reels/TikToks with captions, branded intros/outros, and platform-specific aspect ratios.
### Enterprises → Operations
* **Document processing workflow**: Image → OCR → data extraction → formatting → database entry. Non-technical staff just drop in documents.
* **Customer onboarding automation**: New customer info → personalized welcome video → customized product mockups → email template generation.
## Technical Details
**What gets exposed as app inputs:**
* Text inputs (prompts, descriptions)
* Image uploads (with optional preview)
* Number fields (with min/max constraints)
* Dropdowns/selects (predefined options from model parameters)
* Boolean toggles (yes/no, on/off)
* Sliders (mapped to node parameter ranges)
**What stays hidden:**
* Intermediate nodes (utilities, transformations)
* Model switching logic
* Error handling workflows
* Conditional branching