Team Ai
Modelpublic

mdmachine/ACEStep-XL-Regrind-V1

sourceHugging Facecc-by-nc-sa-4.0updated 2mo agoView on Hugging Face
6likes10kdownloads
Model Card

[image]

ACEStep XL Regrind V1

Oobleck VAE · DiT Layer · Artifact Suppression Stack

Fine-tuned VAE decoder and DiT LoRA targeting the Oobleck hum artifact series at the weight level

Alexander Allan (MDMAchine) · A&E Concepts · CC BY-NC-SA 4.0

![License: CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/) ![Base: ACE-Step XL Turbo](https://huggingface.co/ACE-Step/acestep-v15-xl-turbo) ![VAE: v10b]()


What is this?

ACEStep XL Turbo and the Oobleck VAE produce a characteristic harmonic artifact present in all generations. The artifact has two independent origins:

VAE decoder bias — static reconstruction bias baked into the Oobleck decoder during its original training. Present at consistent amplitude across all generations, independent of length. Characterized via Ableton EQ Eight direct measurement:

FrequencyDepthCharacter
74.4 Hz-1.43 dBSub fundamental, BPM-dependent
94.0 Hz-1.86 dBUpper sub resonance
654–669 Hz-12 to -15 dBPrimary hum, BPM shift ~2Hz/octave
211 Hz~-7 dB3rd harmonic
1090 Hz~-7 dB5th harmonic

DiT body resonance — accumulates over the denoising trajectory, perceptible after ~60 seconds. Hypothesized to arise from ODE manifold shearing in early denoising steps creating resonant modes in FFN activations of the decoder layers.

Both components require independent treatment. Post-processing filters (like the hum notch in MD Audio VAE Tiled) address the symptom. This regrind addresses the cause at the weight level.


Files

VAE (vae/)

FileSizeDescription
vae/acestep_1.5_vae_Regrind_V10b-BF16.gguf337 MBVAE V10b — full-strength regrind (bf16 gguf)
vae/acestep_1.5_vae_Regrind_V10b.safetensors337 MBVAE V10b — full-strength regrind (safetensors)
vae/acestep_1.5_vae_Regrind_V10b_blend50-BF16.gguf337 MBVAE V10b — 50% blend, conservative (bf16 gguf)
vae/acestep_1.5_vae_Regrind_V10b_blend50.safetensors337 MBVAE V10b — 50% blend, conservative (safetensors)
vae/acestep_1.5_vae_Regrind_V9b-BF16.gguf337 MBVAE V9b — full-strength (bf16 gguf, previous gen)
vae/acestep_1.5_vae_Regrind_V9b_blend50-BF16.gguf337 MBVAE V9b — 50% blend (bf16 gguf, previous gen)
vae/acestep_1.5_vae_Regrind_V9b_blend50.safetensors337 MBVAE V9b — 50% blend (safetensors, previous gen)
vae/acestep_1.5_vae_Regrind_V7.safetensors337 MBVAE V7 — legacy (safetensors)

DiT base model (dit/)

FileSizeDescription
dit/acestep_xl_turbo_Regrind_V1.safetensors9.97 GBFull regrinded DiT base model (bf16)
dit/acestep_1.5_xl_turbo_regrind_v1-Q6_K.gguf4.1 GBDiT Q6_K quantized (acestep.cpp / HOT-Step)
dit/acestep_1.5_xl_turbo_regrind_v1-Q4_K_M.gguf2.99 GBDiT Q4KM quantized (acestep.cpp / HOT-Step)
dit/acestep_1.5_xl_turbo_regrind_v1-Q8_0.gguf5.18 GBDiT Q8_0 quantized (acestep.cpp / HOT-Step)

LoRA (lora/)

FileSizeDescription
lora/acestep_xl_turbo_lora_LayerRegrind_V7.safetensors393 MBDiT layer regrind LoRA (strength 0.25–0.50)

Which VAE should I use?

Start with `vae/acestep_1.5_vae_Regrind_V10b_blend50-BF16.gguf` (or .safetensors for ComfyUI) — 50% blend between V10b regrind and the base decoder. Conservative, closest to the base model's original frequency response. Recommended for most material, especially anything spectrally dense or noisy.

Use `vae/acestep_1.5_vae_Regrind_V10b-BF16.gguf` for full-strength V10b suppression when the blend isn't aggressive enough.

V9b variants are the previous generation. Still valid if V10b causes any issues on your material.

V7 safetensors is the oldest format, kept for legacy loaders.


Recommended stack

DiT base:  dit/acestep_xl_turbo_Regrind_V1.safetensors
VAE:       vae/acestep_1.5_vae_Regrind_V10b_blend50.safetensors   (ComfyUI)
           vae/acestep_1.5_vae_Regrind_V10b_blend50-BF16.gguf     (acestep.cpp / HOT-Step)
LoRA:      lora/acestep_xl_turbo_lora_LayerRegrind_V7.safetensors  (strength 0.25–0.50)

On raw ACE-Step XL Turbo without the regrinded base: load the LoRA at 0.50–1.00. No trigger words. Standard ACE-Step prompting.


Pair with MD Audio VAE Tiled

For long-form generation (>30s), pair with the MD Audio VAE Tiled decoder node for ComfyUI. Handles BPM-synced tiling, dual-pass seam elimination, and post-decode spectral cleanup (Wiener denoiser, harmonic preservation crossfade, hum notch stack).

The regrind addresses artifact suppression at the weight level. MD Audio VAE Tiled addresses the decode mechanics. Both together give the cleanest output.

→ [MD Audio VAE Tiled](https://github.com/MDMAchine/MD-Audio-VAE-Tiled)


Related tools

RepoDescription
MD Audio VAE TiledBPM-synced tiled VAE decoder — ComfyUI + C++ + Lua
STORM SamplerAdaptive ODE solver, kills metallic hum at source
MD HAP SchedulerPhysics-derived sigma schedule, HAP+TPT
MD Causal SchedulerCausal timestep scheduler

License

Model weights are licensed under CC BY-NC-SA 4.0.

These are derivative works of ACE-Step XL Turbo (Apache 2.0, ACE Studio / StepFun). Free for personal and research use. Commercial use requires a separate license — contact A&E Concepts.

Source code for associated ComfyUI nodes, C++ headers, and Lua plugins is licensed separately under GPL v3 in the repositories above.


Citation / acknowledgment

If you use these weights in your work, please credit:

Alexander Allan (MDMAchine) — ACEStep XL Regrind V1, A&E Concepts, 2026. https://huggingface.co/mdmachine/ACEStep-XL-Regrind-V1

Base model: ACE-Step v1.5 XL Turbo — ACE Studio / StepFun (Apache 2.0).


© 2026 Alexander Allan (MDMAchine) · A&E Concepts