Chisel
Chisel turns Blender into a real-time SDF modeling studio. Carve with math, not meshes — every surface is mathematically smooth, every boolean blends like molten metal, and the ray-marched viewport shows the final result as you sculpt.
First-run shader compilation
The first time you enable the Chisel render engine in a session, the GPU compiles the ray-marching pipelines on the fly. Expect a noticeable pause on the first viewport frame — typically a couple of seconds on Vulkan, longer on OpenGL. Compilation runs in parallel and is non-blocking, but the viewport will look empty until the first variant is ready.
A few things to know:
- It’s a one-time cost per session. Subsequent renders use the cached pipeline. The compiled binaries are also persisted to disk, so warm starts after the first run are much faster.
- Vulkan is dramatically faster than OpenGL. If your status bar shows the OpenGL backend, switch to Vulkan in Blender’s preferences (
Edit → Preferences → System → Backend) for a 5–10× compile and runtime speedup. - The pie menu warns you. When you press
Qbefore shaders are ready, the menu shows “Shaders not compiled, first run may take longer” — and if you are on the OpenGL backend, “OpenGL backend, Vulkan is much faster”.
If you ever see flat shading, blank silhouettes, or stalls on the first interaction with a new feature (booleans, mirrors, curve primitives), it’s almost always shader compilation finishing in the background. Wait a few seconds and the viewport will catch up.
Why SDFs?
Traditional mesh modeling trades resolution for smoothness. SDFs don’t trade — they compute. A sphere stays spherical at any scale, a blend stays blended through every boolean in the stack, and a rounded corner is a math constant, not a subdivision setting. Production-ready surfaces, no topology headaches.
What’s new in v3 (3.0.0)
- GPU compute renderer — tile binning, primitive culling and tape pruning all run as GPU compute passes; the CPU no longer touches per-tile work
- Two-state render loop — navigation draws synchronously at adaptive resolution while the full-quality frame settles on a background worker and lands when ready
- Dynamic Resolution — automatically scales navigation resolution to hold a target FPS (Performance panel)
- F12 rendering — render the SDF scene through the active camera straight to the Render Result
- Mirror Object — mirror a shape across any other object, not just its own origin; moving the mirror object updates the result live
- Exact depth curves — the Curve primitive now hugs the drawn Bezier exactly: circles stay circular, sharp corners stay sharp, concave spans stay smooth
- Pinned modifiers — pin a modifier to keep it at the bottom of the stack; new modifiers insert above the pins
- RMB menu overhaul — Type picker, Swap Modifier, modal-adjust shortcuts, plus copy / duplicate / snap / parent / collection entries
Ctrl + DChisel duplicate — clone a shape together with its modifier graph (group cuts stay group cuts)- Apply Modifiers — convert to mesh and clean up the source shapes in one click
- Cycle hotkeys —
[/]cycle the primitive type,;/'cycle the modifier operation - Per-chain limits doubled — 256 primitives and 512 instructions per chain
- Apple Silicon native — the new renderer runs on Metal via the arm64 build
For a complete list, see the changelog.
Where to start
- Getting Started — install Chisel, system requirements, first launch
- Primitives — every SDF shape and its parameters
- Modifiers — booleans, mirror, array, emboss, twist, bend, solidify
- Rounding & Blending — the five profiles and when to use each
- Materials — per-object color and the Chisel node editor
- Render Engine — viewport quality, matcap, proxy mesh, baking
- Interface — panels, pie menu, draw tool, gizmos
- Operations — convert to mesh, rebuild, clean, veil