MiniMax H3, Tripo 3D mesh ops & parallel sub-agents
What's new?
New video model — MiniMax H3: MiniMax H3 is now available for text-to-video, image-to-video, and reference-to-video generation at stereo 2K resolution, giving you a high-quality video option with strong motion consistency across all three input modes.
AI controls apply down workspace hierarchy: AI settings configured at the workspace level now propagate down to sub-workspaces automatically, so teams managing multiple workspaces no longer need to configure controls in each one separately.
SVG viewer and Rasterize tool: SVG assets now open in a dedicated viewer with SVG-aware tools, and a new Rasterize option lets you convert any SVG to a flat image in one click when you need it as a raster format downstream.
New 3D model suite — Tripo mesh operations: Six new Tripo operations are now available — Rig, Animate, Retarget, Segment, Remesh, and Retexture — alongside a multi-part mesh viewer. You can now take a generated 3D asset through a full production pipeline without leaving Layer.
Flexible groups: Groups now support configurable roles, usage limits, and member management, giving workspace admins finer control over how teams are organized and what they can access.
Session date filters: You can now filter your session board by date server-side, making it faster to navigate large session histories without loading everything first.
Parallel sub-agent streams: The agent can now run multiple sub-tasks in parallel within a single chat turn, so complex multi-step requests complete significantly faster instead of running sequentially.
Object editor mesh operations: The 3D object editor now supports mesh operations directly, with multi-view INIT slots so you can work with complex assets from multiple angles in the same session.
Aspect ratio picker for edit-only image models: Edit-only image models now expose an aspect ratio picker, giving you output size control that was previously only available on generative models.
Workflow node descriptions: You can now add descriptions to individual nodes inside a workflow, making it easier to document intent and hand off complex graphs to other team members.
Edit workflow metadata outside publish: Workflow name, description, cover image, and node descriptions can now be updated without going through the publish flow, so keeping documentation current doesn't require a full republish.
Canvas tile placement: New tiles generated on the canvas now place beside your current selection rather than at a fixed position, keeping your workspace organised as it grows.
Bug fixes & improvements
Workspace-disabled models hidden from pickers: Models that have been disabled at the workspace level no longer appear in model selection dropdowns, reducing confusion when certain models are intentionally restricted.
Video and mesh frame picks restored: Fixed an issue where selecting a specific frame from a video or mesh as a reference image input was not working correctly.
Hung blueprint runs fixed: Resolved a bug causing blueprint runs to hang indefinitely rather than completing or failing cleanly.
Chat stream recovery: The chat interface now recovers gracefully from mid-turn stream interruptions rather than leaving a generation in an unresolved state.
Library grid performance: Scrolling and loading performance in the asset library grid has been improved for workspaces with large asset counts.
New models11
VideoFrontier video model from Black Forest Labs with native audio, lipsync, and first/last frame control.
VideoGuide FLUX.3 video generation with up to 10 keyframe images pinned across the clip.
VideoMiniMax H3 (Hailuo-03) next-gen open-weight video with native stereo audio at 2K and 24 FPS.
VideoReference-guided MiniMax H3 video at 2K that follows a reference clip shot for shot while the prompt changes the subject or action.
3DHigh-detail Tripo model for production assets with dense geometry and refined textures.
3DRigs a 3D mesh and applies a preset animation from the animation library, returning an animated GLB.
3DGame-ready Tripo model: a single image to clean low-poly 3D with PBR textures.
3DRebuilds a high-poly 3D mesh as a low-poly model at a target face count, baking the original textures onto the result.
3DRetextures an existing mesh from a text prompt or reference image, keeping geometry intact.
3DAuto-rigs a 3D mesh, returning a rigged GLB. Humanoid meshes come back with a Mixamo-named skeleton for direct use in game engines; other body plans use Tripo's own naming, which Mixamo cannot describe.
3DSegments a 3D mesh into parts, returning a part-grouped model for further editing.
Audio
Image
