notapalindrome/ltx25-mlx

MLX split pack of LTX-2.5 for Apple Silicon, for use with dgrauet/ltx-2-mlx / the LTX Video Generator macOS app (ltx2Mlx backend).

You do not need to distill anything. This pack already includes:

File Role
transformer-distilled.safetensors Distilled DiT (8-step, CFG=1)
transformer-dev.safetensors Dev / full DiT
ltx-2.5-22b-distilled-lora-450-bf16.safetensors Distilled LoRA for official Dev two-stage (bundled here so Dev does not need a second gated download)
gemma4-12b-ltx-v1/ + connector.safetensors Gemma 4 12B TE
vae_*, audio_vae, vocoder DiffVAE + conv VAE + audio
spatial_upscaler_*, temporal_upscaler_* Latent ×2 upscalers
duration_head.safetensors Optional duration predictor

Provenance

  • Weights are Derivatives of Lightricks/LTX-2.5 under the LTX-2.x Community License (see LICENSE.md + AUP PDF).
  • MLX layout / conversion follows mlx-community/ltx-2.5-mlx (re-serialization only; no retraining).
  • Distilled LoRA content matches the file published on dgrauet/ltx-2.5-mlx / Lightricks loras/.

App catalog mapping

App model_id How this pack is used
ltx25_distilled --distilled
ltx25_dev --two-stage + transformer-dev + fuse bundled LoRA
ltx25_distilled_ditq8 same pack + overlay notapalindrome/ltx25-mlx-ditq8

Quick CLI

ltx-2-mlx generate \
  --model notapalindrome/ltx25-mlx \
  --distilled \
  --prompt "…" -H 512 -W 768 -f 121 --frame-rate 24 -o out.mp4

Dev two-stage:

ltx-2-mlx generate \
  --model notapalindrome/ltx25-mlx \
  --two-stage --stage1-steps 30 --stage2-steps 3 --cfg-scale 3.0 \
  --distilled-lora ltx-2.5-22b-distilled-lora-450-bf16.safetensors \
  --prompt "…" -o out.mp4
Downloads last month
280
MLX
Hardware compatibility
Log In to add your hardware

Quantized

Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for notapalindrome/ltx25-mlx

Finetuned
(36)
this model