TESSERA
A granular sample processor with a spatial window engine. 2,048 grains of polyphony, stereo spread decided grain by grain, and a scrub mode that treats your sample as terrain instead of a timeline.
Terrain, not timeline.
Every granular processor treats a sample as a strip of tape. TESSERA treats it as landscape. Scrub mode maps the buffer to a two-dimensional plane: horizontal position is time, vertical position is playback window width, and grains are placed on that terrain according to where you tell the cloud to live.
In practice this means one knob performs. Park the cloud low and the module reads wide windows, smooth and tonal. Raise it and the windows narrow, the grains scatter, and the same ten seconds of audio becomes a shimmering field. Because the cloud position is a CV input, the terrain is playable, not just settable.
The spatial engine is the part we are vain about. Each grain's stereo position is decided independently by a correlated random walk, so the cloud breathes as one organism instead of rattling like 2,048 unrelated events. Width is continuous, not a switch. At one extreme the image is a point; at the other it is the room you are sitting in.
- Engine
- Granular, 2,048 grain voices
- Grain length
- 1 ms to 2 s, CV controllable
- Buffer
- 48 seconds stereo at 48 kHz / 32-bit float
- Formats
- WAV, AIFF, FLAC via card slot or USB
- Scrub plane
- Time axis and window axis, both CV inputs
- Spatial engine
- Correlated random walk, per-grain stereo
- Freeze
- Instant, click-free, survives patch changes
- Latency
- 1.4 ms fixed, clocked to the analog outputs
The sheet.
| Format | Eurorack, 18 HP, 26 mm deep including jacks |
|---|---|
| Power | +12 V 420 mA / -12 V 95 mA / 0 mA 5 V measured, full polyphony |
| Architecture | Digital core, analog output stage, honestly labeled as both |
| Grain voices | 2,048 simultaneous, 32-bit float internal processing |
| Sample rate | 48 kHz fixed, 32-bit float buffer, aliasing below -110 dB |
| Dynamic range | 116 dB(A) measured, output stage |
| Inputs | Line in, CV scrub X, CV scrub Y, CV grain length, CV position, CV spread, gate freeze, clock |
| Outputs | Stereo out, per-channel drive send, dry/wet balance |
| Storage | MicroSD slot, 32 banks, sample library on the card |
| USB | USB-C, sample transfer and firmware updates |
| Firmware | 3.2.0, CERN-OHL-S, community grain engines accepted |
| Warranty | 5 years parts and labor, transferable with the serial |
We said the core is analog. TESSERA is the documented exception.
A granular engine is a memory problem, and memory belongs in silicon. Pretending otherwise with an analog charge-bucket architecture would cost three HP of panel and every bit of sound quality for a purity nobody can hear. So the processing core is digital, the output stage is analog, and the panel says so on both blocks.
The rule is not "digital is bad". The rule is that the architecture is printed where you can see it, so you know exactly which part of the signal path is mathematics and which part is physics.
BENCH NOTE / 2026-09
The spatial walk runs on a hardware noise source, not a software random number generator. Two TESSERA units patched identically will never produce identical clouds. We consider this correct behavior for an instrument and incorrect behavior for a computer.
FILED UNDER: TESSERA / SPATIAL ENGINE
“Scrub mode rewired how I think about samples. I stopped slicing and started driving. The CV inputs on both terrain axes make it an instrument, not an effect.”M. OKAFOR / PATTERN & NOISE / REVIEW 2026-02
“The stereo image is the best I have measured in hardware, full stop. The correlated walk means the cloud moves as a body. Nothing else does this.”BENCH TEST / STEREO LAB MONTHLY / ISSUE 41
TESSERA pairs with everything in the catalog.
Especially HELIOS-8: sample its voices at the bench, then rebuild them as terrain. The two modules share the same bus clock and the same calibration language.
What the buffer is for
Forty-eight seconds of stereo at 48 kHz is enough for two full song verses, a field recording of the U8 line from Warschauer to Frankfurter Tor, or ninety seconds of a single cymbal. The buffer is not an empty vessel for samples you will never load. It is a working memory sized for the gestures the module performs: scrub, freeze, stretch, and the micro-loop the envelope finds inside every file.
Load format is WAV, AIFF, or FLAC, up to 32-bit float, from microSD or over USB-C from a computer in mass-storage mode. The file browser is on the module: four markers, one knob, no menus inside menus.
- Buffer
- 48 s stereo, 48 kHz, 32-bit float
- Grains
- 2,048 simultaneous, length 1 ms to 2 s
- Latency
- 1.4 ms, scrub to output
- S/noise
- 116 dB(A), digital core measured at unity
- Formats
- WAV / AIFF / FLAC to 32-bit float
- Storage
- microSD to 512 GB, FAT32 or exFAT
Is the whole path digital?
The grain engine is digital and the I/O stages are analog, measured to 116 dB(A). The honest version: yes, this is a digital module in an analog company, and the technology page's component chapter explains why we think that is the right tool for what it does.
How big can my sample files be?
The buffer holds 48 seconds of stereo at 48 kHz. Files larger than the buffer load their first 48 seconds; the file browser shows the trim and nothing is ever silently truncated. MicroSD up to 512 GB, FAT32 or exFAT from firmware 3.1.0.
Does the crossfield respond to pressure?
Position, not pressure. The scrub field is capacitive for X and Y, and pressure would change what the gesture means, which is the one thing this module refuses to do. CV on both axes is where expression lives.