Instructions to use KoshiMazaki/akuspace-ltx25 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LTX.io
How to use KoshiMazaki/akuspace-ltx25 with LTX.io:
# Install the LTX-2 pipelines git clone https://github.com/Lightricks/LTX-2.git cd LTX-2 uv sync --frozen
# Download the weights from this repo, plus the Gemma text encoder hf download KoshiMazaki/akuspace-ltx25 --local-dir models/akuspace-ltx25 hf download google/gemma-3-12b-it-qat-q4_0-unquantized --local-dir models/gemma-3-12b
# Text/image-to-video with the LoRA on the HQ two-stage base pipeline uv run python -m ltx_pipelines.ti2vid_two_stages_hq \ --checkpoint-path path/to/checkpoint.safetensors \ --distilled-lora path/to/distilled_lora.safetensors 0.8 \ --spatial-upsampler-path path/to/spatial_upsampler.safetensors \ --gemma-root models/gemma-3-12b \ --lora models/akuspace-ltx25/<weights>.safetensors 1.0 \ --prompt "your prompt here" \ --output-path output.mp4 # For image-to-video, add: --image path/to/image.jpg 0 0.8 - Reverb
How to use KoshiMazaki/akuspace-ltx25 with Reverb:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
File size: 2,410 Bytes
4aac74e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 | """Did the Ableton renders pass through a limiter / normaliser?
This decides whether a Python linear crossfade can stand in for an Ableton
render at a lower send level.
If the chain is purely linear (sum of dry + reverb return, no dynamics), then
0.7*dry + 0.3*(dry+reverb) = dry + 0.3*reverb
is EXACTLY an Ableton render at 30% send, and Python is equivalent.
A limiter breaks that. It is level-dependent and non-linear, so a render at 30%
send hits it differently than the 100% render did -- and crossfading afterwards
cannot reproduce that. The tell is a hard ceiling: many files peaking at the
same value rather than scattered.
"""
import wave
from collections import Counter
from pathlib import Path
import numpy as np
DATA = Path("/workspace/Demos/data/acoustic-space-ableton/audio")
def peak_db(p):
with wave.open(str(p), "rb") as w:
n, ch, sw = w.getnframes(), w.getnchannels(), w.getsampwidth()
raw = w.readframes(n)
if sw == 3: # 24-bit
a = np.frombuffer(raw, dtype=np.uint8).reshape(-1, 3).astype(np.int32)
x = (a[:, 0] | (a[:, 1] << 8) | (a[:, 2] << 16))
x = np.where(x & 0x800000, x - 0x1000000, x).astype(float) / 8388608.0
else:
x = np.frombuffer(raw, dtype=np.int16).astype(float) / 32768.0
if x.size == 0:
return None
return 20 * np.log10(np.abs(x).max() + 1e-12)
for label in ("references", "targets"):
peaks = []
for p in sorted((DATA / label).glob("*.wav")):
d = peak_db(p)
if d is not None:
peaks.append(round(d, 2))
if not peaks:
continue
print(f"=== {label}: {len(peaks)} files ===")
print(f" peak range: {min(peaks):.2f} .. {max(peaks):.2f} dBFS")
common = Counter(peaks).most_common(5)
print(" most common peak values:")
for v, c in common:
bar = "#" * min(c, 40)
print(f" {v:>7.2f} dBFS x{c:<3} {bar}")
ceiling = [p for p in peaks if p > -0.5]
print(f" files peaking above -0.5 dBFS: {len(ceiling)} of {len(peaks)} "
f"({100.0*len(ceiling)/len(peaks):.0f}%)")
print()
print("A tight cluster at one value just below 0 => a ceiling was applied")
print("(limiter or normaliser), so the render chain is NOT purely linear and an")
print("Ableton render at lower send will differ from a Python crossfade.")
print("A broad scatter => linear sum, and Python is mathematically equivalent.")
|