← Kyanite Labs · Lab notes

Local image generation, measured: what a small lab's own hardware actually does

Sampler comparison, guidance economics, and co-residency facts from the org's own machines. · 2026-09-28
Limits that travel with this piece: Single-judge limits labeled in-place; all numbers measured on our own hardware.

Draft — Measured results from a consumer APU image stack (working draft v0.1, 2026-09-25)

CS-authored science draft for the Comms publishing pilot. Numbers are our own

receipted measurements; judgments marked CEO-eye are single-judge human calls

and are labeled as such throughout. Limits stated up front: single judge, no

blind scoring, one machine, one model family — the structure (paired seeds,

frozen plans, append-only indexes, hash-stamped outputs) is the defensible

part. This draft makes no claim beyond its n.

1. The setting (why anyone cares)

A ~$1k consumer APU (8-core Strix Halo, iGPU, 80 GiB GTT shared memory) runs a

7B-parameter diffusion transformer (Q4_K) alongside two resident LLM services

(27B champion + 35B MoE) with no partitioning and no mode switching. The

question the org needed answered: what quality and economics does this class

of hardware actually deliver, measured rather than guessed?

2. Measured facts (each with receipt pointers)

Co-residency. Image generation at full quality runs at identical speed

with both LLM residents live (5.84–5.88 s/it across memory modes; peak GTT

58.8 GiB / 80, ~15.5 GiB RAM free). The "park the big models before drawing"

reflex is unnecessary on this memory architecture. [2026-09-23 receipts]

Sampler separation. At 30 steps: samplers euler, euler-a are

speed-identical; dpm++2m costs +5.6% (cfg 4) to +15.9% (cfg 6) wall time. At

low steps (6/10) all samplers are speed-identical — separation is purely a

quality question. CEO-eye judgment: dpm++2m preferred at 30 steps, and is the

only sampler producing usable output at 6 steps; euler-family output at 6

steps is under-integrated (missing content, unresolved exposure). Mechanism

(offered, consistent with solver theory): dpm++2m is a second-order multistep

solver whose per-step error shrinks faster, so it stays on the ODE trajectory

when strides get huge; euler's straight-line steps leave the walk unfinished.

Guidance (cfg) economics. cfg 4 vs 6 at 30 steps: ~10% wall-time savings

(191.7 s vs 211.4 s per 768×512 image, n=10+10). Quality call on real brand

content: pending the CEO eye on like-for-like cells (section 5).

Reload tax. Each fresh CLI invocation pays ~15–20 s model reload; a

20-image batch spends ~6 min of its ~65 min on reloading. Same-process

sequential generation is the first optimization lever (est. ~9% throughput,

larger for short runs). [reload-tax receipts]

Text capability. Stitched/embroidered vertical lettering rendered

perfectly (CEO-eye, both cfg settings) — the hard text class (letters as

thread on fabric). The empty OCR read on the same images is an instrument

artifact, not model failure.

Cost per image. Full-quality 768×512: ~3.2 min wall, ~54 W GPU package

power — hours of generation at quiet-class draw, zero instability across all

receipted runs. Contrast: the same work CPU-only was ~29 min/image and

crashed the machine under sustained load; the GPU lane is both faster and

safer here.

3. The methodology pattern (the reusable part)

Every run in this paper is: a frozen plan (prompts hash-pinned), an

append-only index, per-image sidecars, exact-hash outputs, preflight receipts

gating shared-resource use, and a live served surface for human judgment.

Failures are recorded in place (one cell killed and regenerated on the

record; a terminated-and-adopted run disclosed). This is the paper's second

product: a receipts-first bench pattern other small labs can copy.

4. What we do NOT claim

No blind scoring, no inter-rater statistics, single machine class, single

model. The CEO-eye judgments are directionally strong (paired same-seed

comparisons) and statistically n=1. We claim measured facts and a repeatable

method, not a leaderboard.

5. Open items that complete this draft

  1. CEO keeper calls on the 20-cell real-content set (guidance 4-vs-6

quality; texture-class ceiling verdicts).

  1. The six like-for-like guidance cells (queued in the bench lane) to replace

the prompt-confounded earlier comparison.

  1. Comms: framing, audience, title. CS recommends practitioner-channel

first.

6. Source index

All receipts committed in the lab repo: sampler/low-step grid + reload-tax

run (2026-09-24/25), co-residency receipts (2026-09-23), 20-cell brand

content hunt + telemetry (2026-09-25). Served raw copies available via the

lab's :4690 file surface.