Release notes (0.9.8)
Spin it. Note it. Start ahead.
The graph just learned a third dimension. Generate a 3D model from a prompt or a handful of views, place it, and render perfectly composed references. Leave notes, bend the wires, and get your workflows looking just right using the grid and snap mode. Achieve product consistency, tell the LLM to re-generate that specific variable without loosing context, oh and by the way, it is now vision-enabled. Start your empty workflows from one of the templates and much much more.
2029-09-29: Bumped to 0.9.7 – critical inference queue bug fix
2029-10-02: Bumped to 0.9.8 – new nodes and models, read more below
3D in the workflow
Generate models from prompts, or from image references. Or drop your own 3D models on a 3D input. Orbit, move and scale it against an optional background, and render a still you can send on down the graph.
Notes on the workflow canvas
Leave a note where the graph needs one. Type it, resize it, and snap it to the same grid as the nodes. Notes never run – they just stay put.
A graph more easily read
Edges take the color of what they carry – image, video, audio, or text – so a busy graph is easier to follow. Bend a wire through a waypoint, or set the curve to auto, straight, or square. Done editing a node? Tidy up by switching its view mode to focus on the result rather than the input.
An LLM that can see
The LLM node now has vision. Have it make decisions by looking at things. Generated multiple variables and want to regenerate just one or lock a few and regenerate the rest? No problem! It’s now able to modify its output while staying fully aware of the current context.
Automated subtitles – with control
Add a subtitle layer and generate the cues from a video or audio layer, or from the mix of the whole composition (or simply import a pre-existing SRT). Style the plate – font, alignment, background – and easily edit each cue in the inspector.
Translation documents
Aside from contextually aware ai-driven translations. Compositions now support “Translation document” inputs. Through AI we read and understand your document structure. Feed it with the spreadsheets to have it translate your composition using the document to figure out what goes where. Supports fragmented texts (such as subtitles) and much more.
Quick start from a template
Start from scratch, or use one of the quick start workflow templates that will help you learn and get started. Modify them freely and look at examples on how workflows can be created.
New reframing node (0.9.8)
Place your subject within a customisable canvas and let the model figure out how to extend the frame. Choose between different models.
External editor support (0.9.8)
Need to manually tweak and pixel-push a raster or vector file? No problem, Crust pulls in available editors on your system as well as presenting an option to open in the system default. Crust automatically detects edits and updates accordingly.
Vectorize (0.9.8)
Need to make illustrations into editable curves. Meet the new vectorize node. It allows you to vectorize illustrations to go from raster to curves. Open in your favourite system vector editor through the new external editor features.
Other fixes
- Improved ux for credit cost calculations – improved nested workflow credit display.
- General UX tweaks for a smoother experience.
- Critical video decoding bug fixes for some odd formats.
- Improved color management and customisability.
- Improved export presets and management.
- Registration and validation of .crustpack files.
- Bug fixes and improvements.
Other 0.9.8 updates
- Added new Qwen image, video and language models
- Added MiniMax H3 model
- Further improvements to the workflow grid and snapping sync