Deep house / lounge: generated tracks

Scores: Meta Audiobox Aesthetics, CE = enjoyment, PQ = production quality (1–10).

Stable Audio 3 small (TensorRT, 7× realtime)

sa3_00CE 7.57 · PQ 8.10
sa3_01CE 7.45 · PQ 8.02
sa3_02CE 7.65 · PQ 8.16
sa3_03CE 7.77 · PQ 8.36
sa3_04CE 7.53 · PQ 8.03
sa3_05CE 7.64 · PQ 8.08

Stable Audio 3 medium (staged, experimental)

sa3m_good_00CE 7.03 · PQ 8.04
sa3m_good_01CE 7.24 · PQ 7.92
sa3m_good_02CE 7.52 · PQ 7.92

ACE-Step 1.5 turbo (int8, 2× realtime)

ace_00CE 7.57 · PQ 8.23
ace_01CE 7.52 · PQ 8.30
ace_02CE 6.41 · PQ 7.45
ace_03CE 7.13 · PQ 7.85
ace_04CE 7.80 · PQ 8.35
ace_05CE 7.28 · PQ 8.08

ACE-Step 1.5 sft, 50 steps + CFG (0.33× realtime)

sft_00
sft_01
sft_02
sft_03

Magenta RealTime 2 small (1.4× realtime)

mrt_00CE 7.79 · PQ 8.44
mrt_01CE 5.77 · PQ 7.81
mrt_02CE 7.95 · PQ 8.35
mrt_03CE 6.92 · PQ 6.92
mrt_04CE 7.57 · PQ 8.29
mrt_05CE 7.04 · PQ 8.18

Powered by Stability AI. Other models: ACE-Step 1.5 (MIT), Magenta RealTime 2 by Google (CC BY 4.0).