$ ./tools/sfxgen --prompt 'Single isolated weapon contact against a hard insect carapace: sharp dry chitin crack followed immediately by a hollow shell knock. Brittle, rigid, insect-shell texture with a short natural decay. Material contact only — no weapon swing, no whoosh, no large impact boom, no sub-bass, no music, no voices, no ambience. One impact only near the beginning, followed by silence. Clean immediate attack, dry close sound, reverb tail under 0.3 seconds.' --seconds 0.3 --takes 4 --out audio/source/combat-core/sfx-material-carapace.mp3 # cwd /Users/king/Desktop/Workspace/SRO/zoen # exit 0 Generating take 1/4: /Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/sfx-material-carapace.mp3 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SA3 → MLX text-to-audio ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ prompt 'Single isolated weapon contact against a hard insect carapace: sharp dry chitin crack followed immediately by a hollow shell knock. Brittle, rigid, insect-shell texture with a short natural decay. Material contact only — no weapon swing, no whoosh, no large impact boom, no sub-bass, no music, no voices, no ambience. One impact only near the beginning, followed by silence. Clean immediate attack, dry close sound, reverb tail under 0.3 seconds.' dit sm-sfx decoder same-s σmax 1.00 seconds 1.5s steps 8 seed 2014334959 DiT dtype fp16 cfg 1.0 T_lat 17 (1.58s → trimmed to 1.5s) [1/5] T5Gemma encode ································· 572 ms embeds (1, 256, 768) mlx.core.float16 [2/5] Conditioning ··································· 6 ms cross_attn (1, 257, 768) global (1, 768) [3/5] DiT — load + sample (8 steps, σmax=1.00) load 0.5s schedule 1.000 · 0.994 · 0.984 · 0.958 · 0.891 · 0.746 · 0.512 · 0.274 · 0.000 sample 873 ms (109 ms/step) [4/5] Decoder (same-s, FP32) load 92 ms decode chunked (chunk=8, ovl=2) → 32 ms patches (1, 512, 272) [5/5] Unpatch + write WAV ···························· 3 ms audio (2, 66150) peak 6.981 rms 0.289 Peak RAM by stage T5Gemma encode ███████████············· 708 MB Conditioning █████████··············· 539 MB DiT load ██████████████·········· 878 MB DiT sample ████████████████████████ 1.45 GB ← max Decoder load ███····················· 210 MB Decode ███████················· 438 MB Unpatch + WAV ███····················· 211 MB ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ done 2.11s wall → 1.5s audio → 0.71× realtime peak RAM 1.45 GB seed 2014334959 ▸ saved /Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/.sfxgen-_mltttfa/source.wav ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ {"path": "/Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/sfx-material-carapace.mp3", "sample_rate": 44100, "channels": 2, "seconds": 0.3, "peak_dbfs": -12.869, "bytes": 7715, "boundary_jump": 0.0011888178414665163} Generating take 2/4: /Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/sfx-material-carapace_2.mp3 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SA3 → MLX text-to-audio ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ prompt 'Single isolated weapon contact against a hard insect carapace: sharp dry chitin crack followed immediately by a hollow shell knock. Brittle, rigid, insect-shell texture with a short natural decay. Material contact only — no weapon swing, no whoosh, no large impact boom, no sub-bass, no music, no voices, no ambience. One impact only near the beginning, followed by silence. Clean immediate attack, dry close sound, reverb tail under 0.3 seconds.' dit sm-sfx decoder same-s σmax 1.00 seconds 1.5s steps 8 seed 1474876316 DiT dtype fp16 cfg 1.0 T_lat 17 (1.58s → trimmed to 1.5s) [1/5] T5Gemma encode ································· 208 ms embeds (1, 256, 768) mlx.core.float16 [2/5] Conditioning ··································· 2 ms cross_attn (1, 257, 768) global (1, 768) [3/5] DiT — load + sample (8 steps, σmax=1.00) load 0.2s schedule 1.000 · 0.994 · 0.984 · 0.958 · 0.891 · 0.746 · 0.512 · 0.274 · 0.000 sample 737 ms (92 ms/step) [4/5] Decoder (same-s, FP32) load 48 ms decode chunked (chunk=8, ovl=2) → 32 ms patches (1, 512, 272) [5/5] Unpatch + write WAV ···························· 2 ms audio (2, 66150) peak 6.437 rms 0.350 Peak RAM by stage T5Gemma encode ███████████············· 708 MB Conditioning █████████··············· 539 MB DiT load ██████████████·········· 878 MB DiT sample ████████████████████████ 1.45 GB ← max Decoder load ███····················· 210 MB Decode ███████················· 438 MB Unpatch + WAV ███····················· 211 MB ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ done 1.26s wall → 1.5s audio → 1.19× realtime peak RAM 1.45 GB seed 1474876316 ▸ saved /Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/.sfxgen-s6576ouo/source.wav ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ {"path": "/Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/sfx-material-carapace_2.mp3", "sample_rate": 44100, "channels": 2, "seconds": 0.3, "peak_dbfs": -10.257, "bytes": 7140, "boundary_jump": 0.0024588359519839287} Generating take 3/4: /Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/sfx-material-carapace_3.mp3 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SA3 → MLX text-to-audio ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ prompt 'Single isolated weapon contact against a hard insect carapace: sharp dry chitin crack followed immediately by a hollow shell knock. Brittle, rigid, insect-shell texture with a short natural decay. Material contact only — no weapon swing, no whoosh, no large impact boom, no sub-bass, no music, no voices, no ambience. One impact only near the beginning, followed by silence. Clean immediate attack, dry close sound, reverb tail under 0.3 seconds.' dit sm-sfx decoder same-s σmax 1.00 seconds 1.5s steps 8 seed 1526134583 DiT dtype fp16 cfg 1.0 T_lat 17 (1.58s → trimmed to 1.5s) [1/5] T5Gemma encode ································· 209 ms embeds (1, 256, 768) mlx.core.float16 [2/5] Conditioning ··································· 2 ms cross_attn (1, 257, 768) global (1, 768) [3/5] DiT — load + sample (8 steps, σmax=1.00) load 0.2s schedule 1.000 · 0.994 · 0.984 · 0.958 · 0.891 · 0.746 · 0.512 · 0.274 · 0.000 sample 799 ms (100 ms/step) [4/5] Decoder (same-s, FP32) load 47 ms decode chunked (chunk=8, ovl=2) → 33 ms patches (1, 512, 272) [5/5] Unpatch + write WAV ···························· 1 ms audio (2, 66150) peak 7.703 rms 0.543 Peak RAM by stage T5Gemma encode ███████████············· 708 MB Conditioning █████████··············· 539 MB DiT load ██████████████·········· 878 MB DiT sample ████████████████████████ 1.45 GB ← max Decoder load ███····················· 210 MB Decode ███████················· 438 MB Unpatch + WAV ███····················· 211 MB ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ done 1.33s wall → 1.5s audio → 1.13× realtime peak RAM 1.45 GB seed 1526134583 ▸ saved /Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/.sfxgen-no0_qx35/source.wav ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ {"path": "/Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/sfx-material-carapace_3.mp3", "sample_rate": 44100, "channels": 2, "seconds": 0.3, "peak_dbfs": -11.903, "bytes": 7506, "boundary_jump": 0.002818408072926104} Generating take 4/4: /Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/sfx-material-carapace_4.mp3 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SA3 → MLX text-to-audio ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ prompt 'Single isolated weapon contact against a hard insect carapace: sharp dry chitin crack followed immediately by a hollow shell knock. Brittle, rigid, insect-shell texture with a short natural decay. Material contact only — no weapon swing, no whoosh, no large impact boom, no sub-bass, no music, no voices, no ambience. One impact only near the beginning, followed by silence. Clean immediate attack, dry close sound, reverb tail under 0.3 seconds.' dit sm-sfx decoder same-s σmax 1.00 seconds 1.5s steps 8 seed 1293860482 DiT dtype fp16 cfg 1.0 T_lat 17 (1.58s → trimmed to 1.5s) [1/5] T5Gemma encode ································· 208 ms embeds (1, 256, 768) mlx.core.float16 [2/5] Conditioning ··································· 2 ms cross_attn (1, 257, 768) global (1, 768) [3/5] DiT — load + sample (8 steps, σmax=1.00) load 0.2s schedule 1.000 · 0.994 · 0.984 · 0.958 · 0.891 · 0.746 · 0.512 · 0.274 · 0.000 sample 838 ms (105 ms/step) [4/5] Decoder (same-s, FP32) load 48 ms decode chunked (chunk=8, ovl=2) → 36 ms patches (1, 512, 272) [5/5] Unpatch + write WAV ···························· 1 ms audio (2, 66150) peak 4.412 rms 0.243 Peak RAM by stage T5Gemma encode ███████████············· 708 MB Conditioning █████████··············· 539 MB DiT load ██████████████·········· 878 MB DiT sample ████████████████████████ 1.45 GB ← max Decoder load ███····················· 210 MB Decode ███████················· 438 MB Unpatch + WAV ███····················· 211 MB ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ done 1.37s wall → 1.5s audio → 1.10× realtime peak RAM 1.45 GB seed 1293860482 ▸ saved /Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/.sfxgen-4ierfy45/source.wav ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ {"path": "/Users/king/Desktop/Workspace/SRO/zoen/audio/source/combat-core/sfx-material-carapace_4.mp3", "sample_rate": 44100, "channels": 2, "seconds": 0.3, "peak_dbfs": -3.198, "bytes": 7716, "boundary_jump": 0.0042316397884860635}