LTX-2.5 (22B Distilled Transformer) - GGUF Quantizations

This repository contains GGUF quantizations for Lightricks/LTX-2.5, including the 22B distilled transformer model.

These weights are compatible with ComfyUI-GGUF.


Converted Files & Models

1. Main Diffusion Models (diffusion_models/)

  • Base Architecture: LTX-2.5 22B Distilled Transformer
  • Directory: ComfyUI/models/diffusion_models/ (or ComfyUI/models/unet/)
File Quant Type Size Notes
ltx-2.5-22b-distilled-transformer-Q4_0.gguf Q4_0 ~11.65 GB Optimized for 16GB–24GB VRAM GPUs
ltx-2.5-22b-distilled-transformer-Q8_CR.gguf Q8_CR ~20.04 GB High precision compressed-row quant
ltx-2.5-22b-distilled-transformer-Q8_0.gguf Q8_0 ~21.22 GB Near FP16/BF16 quality, fits on 24GB+ VRAM

Example ComfyUI Workflows (workflows/)

This repository includes ready-to-use ComfyUI workflow JSON files in the workflows/ folder:

  1. Text-to-Video (T2V):
    • File: workflows/video_ltx2_5_t2v-gguf.json
    • Generates video directly from text prompts using the GGUF transformer.
  2. Image-to-Video (I2V):
    • File: workflows/video_ltx2_5_i2v-gguf.json
    • Animates a single source frame with synchronized audio and motion.
  3. First & Last Frame to Video (FLF2V):
    • File: workflows/video_ltx2_5_flf2v-gguf.json
    • Interpolates seamless video transitions between two keyframe images.

ComfyUI Setup & Requirements

Required Custom Nodes

  • GGUF Loader: molbal/ComfyUI-GGUF (Registry: comfyui-gguf-reboot)
    • Node: UnetLoaderGGUF (select the .gguf transformer model)

Additional Model Dependencies

Ensure you place the upstream support weights into their respective folders:

  • Text Encoder / CLIP: Gemma 4 12B with projection (ComfyUI/models/clip/ or text_encoders/)
  • Video VAE: ltx-2.5-video-vae-bf16.safetensors (ComfyUI/models/vae/)
  • Audio VAE: ltx-2.5-audio-vae-bf16.safetensors (ComfyUI/models/vae/)
Downloads last month
6,990
GGUF
Model size
21B params
Architecture
ltxv
Hardware compatibility
Log In to add your hardware

4-bit

8-bit

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