Updates

This release adds 4K resolution support for Kling video editing, expands the Video Compositor with per-layer cropping and scaling modes, and improves element support in Kling workflows.

Native Kling API with 4K video editing

The Kling O3 Video Edit node now connects directly to Kling’s API 2.0 instead of using a fal.ai wrapper. This unlocks 4K resolution for video editing tasks — previously capped at 1080p. The Quality selector now includes Standard (720p), Pro (1080p), and 4K options, consistent with other Kling O3 generation nodes.

Existing workflows automatically migrate to the new node type while preserving quality settings. The native integration also supports the full Kling content specification, including base_video editing with proper audio handling constraints.

Video Compositor: per-layer crop editor

Add precise cropping to individual layers in the Video Compositor with a new on-canvas crop editor. Click the crop mode toggle in the inspector to reveal corner and edge handles that let you resize the crop rectangle while seeing the full source content dimmed outside the selection. The crop is applied before transformation, so rotated layers maintain correct cropping behavior.

Cropped content re-centers on the layer’s anchor point, and the transform gizmo, scaling presets, and size readout all operate on the cropped region rather than the full source.

Video Compositor: scaling modes and resize handles

Layers in the Video Compositor now support independent X/Y scaling with Fit, Fill, and Stretch presets. Toggle aspect ratio linking on or off to control whether scaling remains uniform or allows free distortion. Eight resize handles on the transform gizmo (corners and edges) let you interactively scale layers while maintaining the opposite handle’s position.

A live size readout shows the rendered dimensions in pixels, and quick-rotate buttons (+90°/–90°/reset) provide convenient rotation controls beside the rotation field.

Quality of life

• Upload progress: File uploads now show real byte-level progress on R2 storage instead of simulated ramps
• Kling references: Prompt autocomplete correctly inserts @image_1/@element_1 tags that match Kling’s API requirements
• Gaussian splat detection: Uploaded .ply files are automatically detected as splats or meshes based on their actual content
• Kling pricing: Video edit nodes now correctly apply quality-tier multipliers for Standard/Pro/4K resolutions

This release introduces a powerful new Video Compositor for client-side video layering and adds MiniMax H3 video generation. The Image Compositor also gets a clearer label to distinguish it from its video counterpart.

Video Compositor node

Add a Video Compositor node to layer multiple video and image inputs into a single MP4 with AAC audio—all processed locally in your browser without using credits. The fullscreen editor provides a timeline with frame-accurate scrubbing, per-layer controls for blend modes, opacity, transforms, trimming, and positioning. Adjust output settings like resolution, duration, and frame rate before rendering. The compositor handles audio mixing from visible video layers and includes transport controls for previewing your edits.

MiniMax H3 video generation

Generate high-quality videos with MiniMax H3, now available via WaveSpeed. This model produces native 2K resolution video with stereo audio from text prompts, images, or reference videos. Find it under Video > MiniMax in the node menu.

Quality of life

• Image Compositor node renamed from “Compositor” for clarity
• Project names update immediately in switchers after renaming
• Asset browser grid scrolls independently, preventing overscroll bounce on toolbar and panels

Add markdown notes to your canvas and restore your preferred view when switching workflows. This release also brings advanced controls to Luma Ray 3.2 video-to-video jobs and improves asset management.

Markdown Notes

Add free-floating markdown notes to your canvas for documentation and collaboration. Notes are non-executable nodes that never enter the compute graph—perfect for explaining complex workflows to teammates. They support standard markdown formatting and maintain a sticky-note appearance with resizable dimensions.

Canvas View Memory

The canvas now remembers your viewport position and zoom level for each workflow. When switching between projects, you’ll return to exactly where you left off, eliminating the disorientation of landing on empty space. The view state is stored per-device and adapts to your current window size.

Luma Ray 3.2 Advanced Controls

Video-to-video jobs with Luma Ray 3.2 now expose per-signal conditioning controls and keyframe anchoring. Fine-tune pose, depth, normals, and trajectory signals individually, or pin guide images to specific frames using the new keyframe system. These advanced options appear when you set an explicit edit strength rather than using auto mode.

Unified Asset Selection

Asset selection now uses a single consistent model across the library. Selecting an asset automatically opens the details panel, while multi-selection shows bulk actions. The previous separate active item and multi-select states have been unified for a more intuitive experience.

Quality of life

• Media load errors now suggest checking VPN settings when CDN access is blocked
• Image uploads are capped at fal’s 100MB limit with clear error messages
• Success toasts removed for actions with clear visual feedback (favorites, uploads, dialog closures)
• Gated parameters now properly disable on both canvas and jobs surfaces
• 3D models and splats can be dragged into folders like other media types

This release sorts project favorites inline with the rest of your projects, creating a consistent experience across the app.

Quality of life

• Favorite projects in the Projects grid now appear sorted inline with other projects instead of in a separate pinned section at the top.

This release introduces a redesigned asset library toolbar and improves language detection to prevent crashes from browser translation features.

Redesigned asset library toolbar

The asset library now features a refreshed toolbar with improved organization and clearer visual hierarchy. Sort options have been consolidated into a field picker with an attached direction toggle, while layout controls like thumbnail fit and card size have moved to a dedicated “View” popover. The toolbar adapts to different screen sizes, stacking vertically on narrow displays and splitting into two connected sections when space allows.

Improved language detection

The app now properly detects browser language preferences and serves localized interfaces when available, preventing crashes that occurred when browsers attempted to translate the page. Users with browsers set to Japanese, French, Spanish, or Chinese will automatically see the app in their preferred language when those translations are available.

This release adds a new generative splatting node and resolves several workflow synchronization issues.

TripoSplat single-image generative splatting

A new Gaussian Splat node lets you generate 3D splats from a single image using TripoSplat’s generative flow-matching model. Unlike reconstruction pipelines that require multiple input images, this creates a 262K-gaussian 3D scene in 1–2 minutes from just one photo. The node includes advanced diffusion controls for steps, guidance scale, flow shift, and seed parameters.

Quality of life

• Project renaming and sharing no longer trigger version conflict dialogs when editing the same workflow
• Failed jobs now properly propagate their status to the canvas instead of showing indefinite “Generating” states
• Branch preview environments handle user provisioning and billing identities automatically

This release adds reference image, video, and audio support to the WaveSpeed Seedance 2.0 node, matching the capabilities available through the fal provider.

Reference support for WaveSpeed Seedance 2.0

The WaveSpeed Seedance 2.0 node now accepts reference images, videos, and audio inputs alongside text prompts. You can now supply up to 9 reference images, 3 reference videos (totaling ≤15 seconds), or 3 reference audio clips (≤15 seconds) to guide video generation.

When reference inputs are connected, the node automatically routes to the text-to-video endpoint and intelligently handles frame references—connected Start/End frames are folded into the reference images array with proper positional tagging. The prompt widget includes autocomplete support for reference tags (@Image, @Video, @Audio), making it easy to reference specific inputs within your text prompt.

This release introduces community workflow sharing, batch connections from multi-node selections, and drag-to-select marquee in the asset library. You can now also track spending by project and download large asset collections more efficiently.

Submit workflows to the community showcase

You can now submit your own workflows to the public template gallery for others to use. From the workflow menu, select “Send to public showcase” to open a submission dialog where you can provide a name, category, and optional description. Submitted workflows undergo admin review before appearing in the gallery, and approved templates are credited with a “by {name}” attribution.

Batch-drag connections from multi-node selections

When you select multiple nodes that share the same output type, a virtual output handle appears on the selection’s bounding box. Dragging this handle lets you connect all selected nodes to a compatible multi-input target in one gesture—either dropping directly on a node that accepts multiple inputs or opening an add-node menu filtered for batch-compatible options. The connections are wired in top-to-bottom, left-to-right order with visible numbering during the drag.

Drag-to-select marquee in the asset library

The asset library grid now supports drag-to-select with a marquee tool. Click and drag over empty space or unselected tiles to make a selection—hold Shift to add to the current selection, Alt/Ctrl/Cmd to subtract, or use no modifier to replace. The marquee mode updates live as you change modifiers mid-drag, and auto-scrolls when dragging near the top or bottom of the scroll container.

Track spending by project

The Activity tab in your account settings now includes a “By project” view that shows credit spend aggregated per project. This helps users managing multiple projects for different clients understand their spending distribution across workflows. The view includes generation counts, credit totals, and EUR equivalents for each project.

Quality of life

• On the canvas: Drop 3D models and Gaussian splats directly onto the canvas—they’ll create upload nodes with correct media type detection and previews.
• In the asset library: Create folders inline from the sidebar with an optimistic insert that appears instantly.
• For downloads: Bulk downloads now generate server-side zips, removing client memory limits and raising the maximum download size to 3 GB.
• For new users: The empty-canvas video starter preset now uses lower-cost settings to ensure first-generation success on Free tier accounts.

This release adds Lucy Edit 2.5 for video editing, expands batch download limits with smarter memory-based gating, and introduces four new language options. The image iterator now supports larger batches and includes new inspection tools.

Lucy Edit 2.5

A new video-to-video editing node powered by Decart’s Lucy Edit 2.5 model is now available. It edits source videos using text prompts and optional reference images, joining the existing Lucy Restyle node. The model runs through Decart’s native API with pricing based on input clip duration.

Smarter batch downloads

Batch download limits have been raised from 10 to 50 files, now governed by a 1 GB total size limit rather than just file count. This prevents memory issues while allowing larger downloads of smaller files. The system automatically handles mixed selections with known and unknown file sizes, and single large files download natively without size restrictions.

Expanded language support

The interface now supports Spanish, French, Simplified Chinese, and Japanese alongside English. Language selection is available in your preferences, with date formatting and full UI translation for each locale.

Enhanced image iterator

The image iterator now supports up to 50 images per batch and includes a grid view with right-click actions for each image. Downstream nodes clearly indicate when they’re processing iterator outputs versus combined images.

Quality of life

• Seedance 2.0 now properly uses connected first/last frames when reference images are present
• Video nodes use consistent “Start Frame” and “End Frame” labeling
• Poll-based job completion is more reliable for long-running tasks
• Seedream 4 and 4.5 image nodes are deprecated in favor of Seedream 5

This release introduces a new high-quality image generation model and expands editing capabilities for 3D Gaussian splats. You can now create 3D scenes directly from text prompts and edit existing splats non-destructively.

Bytedance Seedream 5 Pro

A new flagship image generation model is now available, offering higher quality outputs than the existing Seedream 5 Lite tier. The unified image node automatically routes to the appropriate endpoint based on whether you provide image inputs—text-to-image generation when no images are connected, or image editing when inputs are present. Pricing is based on output image area, with the minimum size placing generations in the higher-quality tier.

Edit Gaussian Splats

Gaussian splat outputs now include an “Edit Splat” action that opens a dedicated editor in a modal window. This non-destructive workflow lets you modify splats and save them as new Import nodes, preserving the original source. The editor exports in the SPZ format, ensuring compatibility with the existing preview and rendering pipeline.

3D Scene Generation from Text

Create 3D scenes directly from text prompts through a new integration accessible from two entry points: the reference video picker in canvas parameter panels and runs launcher, and the empty canvas preset gallery. The generated flythrough videos return as standard video assets ready for use in your projects.

Quality of life

• Machine translations updated across the interface