Documentation

Write a text prompt describing what you want to create. Select a model from the dropdown at the top of the prompt panel - each model displays its name, provider, and supported features. Choose an aspect ratio (1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3) and click Generate. Results appear in your feed.

The catalog includes Gemini Imagen, Flux Pro, Flux Dev, Flux Schnell, DALL-E 3, GPT Image, Stable Diffusion XL, Ideogram, Recraft, Midjourney, and more. Each model has different strengths - photorealism, illustration, text rendering, speed, etc. The model dropdown shows which features each model supports.

Drag and drop an image directly onto the prompt area, paste from clipboard (Cmd+V / Ctrl+V), or click the attachment button to upload. Reference image limits depend on the selected model, and the prompt area displays only the slots that model supports. The AI uses your references as visual guides - matching colors, style, composition, or specific elements depending on the model.

Use the batch selector (1x, 2x, 4x, 8x) to generate multiple variations from the same prompt simultaneously. Each variation uses the same settings but produces a unique interpretation. This is ideal for exploring different directions quickly before committing to a style.

Click the variation button on any generated image to create alternative versions. The AI uses your original image as a starting point and generates new interpretations while maintaining the core composition and style. This is different from regenerating - variations are intentionally similar to the source.

For supported models, the camera style dropdown lets you specify shot types: close-up, medium shot, wide shot, extreme wide, aerial/drone, low angle, high angle, over-the-shoulder, POV, Dutch angle, and more. This gives you cinematic framing without needing to describe it in your prompt text.

Toggle the enhance button (sparkle icon) to let AI rewrite your prompt before generation. Enhancement adds photographic detail, composition guidance, lighting descriptions, and model-specific optimizations automatically. The enhanced prompt is shown so you can learn what works well with each model.

Describe what you want changed in natural language - no masking required. The AI interprets your instructions and applies edits across the entire image. Examples: "Make it nighttime", "Add snow on the mountains", "Change the color of the car to red".

Paint a mask over the area you want to change, then describe what should appear there. Adjust brush size with [ and ] keys. Use add/subtract mask modes for complex selections. The AI generates content that blends seamlessly with the surrounding area.

Paint over any object to remove it. The AI fills in the background naturally, maintaining the scene coherence. Works best with objects that have clear boundaries against their background.

One-click background removal extracts the subject cleanly onto a transparent background. Useful for compositing, product photography, and creating assets for other projects.

Rotate and move the virtual camera in 3D space to adjust perspective. This creates parallax and depth effects from a single 2D image. Control pitch, yaw, and distance to reframe your shot without regenerating.

Crop with preset aspect ratios or custom dimensions. The crop tool supports standard ratios (1:1, 16:9, 4:3, etc.) and freeform cropping.

Enhance resolution using AI upscaling models. Increase image dimensions while preserving and enhancing detail. Supported models can expose scale, format, bit-depth, compression, and TIFF output settings for higher-end delivery.

Outpainting exists as an experimental editing capability in the model constraint layer, but it is not exposed in the default editor toolbar yet. For production work, use crop, instruction editing, and inpainting until a dedicated outpaint control is available.

Describe the scene you want in your prompt. The AI generates a video clip matching your description. Most models produce 5-10 second clips at 720p-1080p. Use AI prompt enhancement for better cinematic results.

Use any generated or uploaded image as the first frame. Drag an image onto the prompt area, paste from clipboard, or click the start frame slot. Some models also support an end frame for precise shot transitions - the AI interpolates between your two frames.

Leading video models: Sora, Veo 3.1, Kling 2.0, Runway Gen-4, Seedance, Hailuo Minimax, Luma Dream Machine, Pika, and more. Each model has different strengths in motion quality, prompt adherence, and style.

Extend any generated video to add more frames. The AI continues the motion and scene from the last frame. Works best when the final frame has clear motion direction. You can extend multiple times to build longer sequences.

Upscale any video to 4K resolution using Topaz AI. The upscaler enhances detail and sharpness dramatically. Supports 10-bit HEVC (Main10) profile for HDR workflows. Also available as 2K intermediate upscale.

Choose clip length per model - typically 5s or 10s. After generation, adjust playback speed with slow-motion or speed-up controls. Speed changes are non-destructive and can be applied in the Story editor.

Generate natural-sounding speech with custom voice selection. Choose from dozens of preset voices or clone your own. Control emotion, pacing, and style per voice provider.

Describe the mood, genre, and instrumentation you want. Generate background tracks for films, ads, or social content in any duration. Supports multiple music generation models.

Create foley SFX from text descriptions. Great for ambient sounds, impacts, transitions, and environmental audio for video projects. Can also generate SFX from video input on supported models.

Upload a short audio sample and the AI learns the voice. Then generate new speech in that voice with any text. Multiple providers available including ElevenLabs, Hume Octave, and Fish Audio.

Split any audio track into individual stems: vocals, drums, bass, and other instruments. Useful for remixing, isolating dialogue, or creating karaoke versions.

Convert speech to text with speaker diarization (identifying who said what). Supports multiple languages and outputs timestamped transcripts.

Place sticky notes (8 color presets: yellow, pink, blue, green, orange, purple, red, gray), text labels with customizable fonts and weights, images/video/audio embeds, and shapes (rectangle, circle, diamond, star, arrow). Each element can be repositioned, resized, and styled independently.

Press T for the text tool, R for shape (rectangle), L for connector (line) tool, and N for sticky note. Press Escape to cancel the active tool or deselect all elements. Click on the canvas to place the element. For shapes, drag to set the size.

Draw arrows between any two elements by activating the connector tool (L key) and dragging from one element to another. Connectors follow the elements when they are moved, creating flow charts and relationship diagrams.

Use Add Media to choose Upload File, Project Files, or Generate Image. Upload File places local media on the canvas, Project Files lets you browse existing workspace assets, and Generate Image creates a new image at the current viewport location.

Drag images, videos, or audio files directly from your file system onto the canvas. You can also paste media from the clipboard (Cmd+V). Dropped files are uploaded and placed at the drop location. Drag media from the Project Files sidebar to place it.

Use rectangle, circle, diamond, star, arrow, and connector tools for layout and flow diagrams. Shapes support color, fill, stroke, size, and rotation changes so boards can carry visual hierarchy without leaving the canvas.

Select multiple elements, then press Cmd+G to group them into a frame. Grouped elements move together. Press Cmd+Shift+G to ungroup. Right-click for additional context menu options.

Export your canvas as a PDF or image for presentations and sharing. The export captures all visible elements at high resolution.

Projects are organized into scenes, and scenes contain individual shots. Each shot has its own prompt, model selection, and generated media. Reorder shots by dragging, add new scenes to organize narrative beats, and collapse scenes to focus on specific parts of your project.

Define characters in the character library with a name, description, and reference images. Use @mention tagging in shot prompts to reference characters - the AI maintains visual consistency across all shots that reference the same character. Add location tags for environmental consistency.

Generate images or videos for each shot individually. For video shots, use start frame control to ensure smooth transitions between shots. Toggle auto-generate to batch create media across all shots in a scene automatically.

The timeline shows all shots sequentially with trim sliders for each. Adjust per-shot speed, add speed ramps, link start or end frames across neighboring shots, navigate between shots with previous/next controls, and preview the full sequence with play/pause. The timeline updates in real time as you generate new media.

Add transitions at shot boundaries to smooth scene changes. Use transitions for fades, dissolves, morphs, and other temporal effects while preserving each shot as its own editable unit.

Add text, image, or video overlays on layered tracks. Text overlays support presets, font styling, stroke, background, shadows, and text animation. Media overlays support position, size, rotation, opacity, border, shadow, trim, playback speed, volume, and keyframes.

Add up to 5 concurrent audio tracks. Browse your audio library, generate music/SFX/speech inline, or upload audio files. Each track can be positioned independently on the timeline, trimmed, faded, assigned to a channel, volume-automated with keyframes, or pinned to a shot with a relative offset.

Apply AI enhancements directly from the editor: upscale video to higher resolution, add lip sync from an audio file, change voices with ElevenLabs/Hume/Fish Audio, generate SFX, or run any saved workflow on selected clips.

Export your completed project as high-quality MP4 with optional 4K Topaz upscaling, a zipped Sequencer project, CapCut project files for further editing, or Premiere Pro XML for professional post-production workflows.

Before delivery, run review workflows to inspect timing, visual issues, and edit quality. Compliance tools can analyze export IP risk and generate reports for production handoff.

Press Tab anywhere on the canvas to open the Quick Add menu at your cursor position. The menu shows all available node types grouped by category (Image, Video, Audio, 3D, Text, Logic, I/O, Effects). Search by typing to filter. Press Enter to add the selected node. You can also right-click the canvas to open the full node palette.

Drag from an output socket (right side of a node) to an input socket (left side of another node) to create a connection. Sockets are color-coded by type: string, number, boolean, image, video, audio, 3D, etc. Only compatible types can be connected. Connections are drawn as curved lines showing data flow direction.

Drag from an output socket into empty space to open the Quick Add menu, pre-filtered to show only nodes with compatible input sockets. Select a node and it will be created and automatically connected to the output you dragged from. This is the fastest way to build a chain of nodes.

Image nodes: Generate, Inpaint, Erase, Variations, Variation Grid, Depth Estimation, ControlNet, Composition, Image Filter, Convert Image, Paint Mask, Color Grade, Transform. Video nodes: Generate, Extend, Trim, Stitch, Split, Speed, Replace Audio, Upscale, Timeline Composite, Timeline Editor, Lip Sync, Pad Media, Convert Video, Extract Audio, Extract Image. Audio nodes: Generate Audio, Video to SFX, Separate Audio, Mix Audio, Transcribe, Split Audio, plus hidden internal Clone Voice and Generate Voice nodes. 3D nodes: Image-based Generate 3D, Render 3D, 3D Model constant. Text nodes: LLM Text, Text constant, File constant, String Join, Search & Replace, Separate Text, Raw Output. Logic/API nodes: Number, Boolean, Transform constant, Math, Compare, Select Index, Pack Array, Pluck Array, Map Entities, Reroute, HTTP Request, File Download, Output, Media Viewer, Sticky Note, deprecated Endpoint Start and Test Endpoint. Enterprise: Train LoRA.

The workflow reference in docs/references/workflow-node-reference.md is generated from components/workflows/data/nodeDefinitions.ts. Regenerate or review that file whenever node definitions, visibility flags, enterprise flags, categories, socket names, or descriptions change.

Select multiple nodes and group them together with a labeled colored frame. Groups help organize complex workflows visually. Each group has a customizable label and color. Nodes within a group can still be individually selected and configured.

Click Run to execute the workflow. Nodes execute in dependency order - upstream nodes complete before downstream nodes start. Each node shows its status: idle, pending, running, completed, or error. Errors display inline with the failed node. The debug panel shows execution details and timing. A watchdog automatically detects stuck nodes and marks them as timed out.

Click Publish to save your workflow as a reusable API endpoint. Enable the endpoint, copy the Endpoint ID, and call it via REST API from any application. Published workflows can be shared publicly in the gallery. Viewers can clone published workflows into their own account if the creator allows it.

Nodes that support batch execution can process multiple inputs in parallel. The batch status shows total items, completed count, and individual results. Failed items within a batch do not block other items from completing.

Use an image from the media library, image generation, upload flow, or workflow graph as the input for 3D generation. Text-only 3D generation is not exposed in the current public tool surface.

The workflow editor includes Generate 3D, Render 3D, and 3D Model constant nodes. Generate 3D accepts an image input, Render 3D accepts a model plus camera controls, and 3D Model lets you reference an uploaded GLB, OBJ, or FBX-style asset.

Generated 3D media can be downloaded, viewed in the browser, or imported through DCC integrations such as the Blender and Houdini plugins when the file format is supported by that host application.

Enhance existing video media with upscale, denoise, sharpen, and HDR-style processing where supported by the selected model. Use upscale for final delivery or as a higher-quality intermediate before editing.

Use a video plus an audio file or URL to queue a lip sync job. The output is saved as a new media version so the original clip remains available.

Generate or replace video audio from a prompt, create SFX for an existing clip, or run voice change to produce a new voice-styled media version.

Run any enabled workflow API endpoint against the selected video when a reusable graph already contains the processing steps you need. Use this for studio-specific transforms, custom review passes, or multi-step processing chains.