LTX-2.5 22B — Comfy - GGUF
GGUF quantizations of Lightricks' LTX-2.5, converted for efficient execution in tools like ComfyUI. Example video is rendered using Q4_K_M quantize variant.
Model Explanation
LTX-2.5 is an open-world model with open weights, built for local execution and fine-tuning. It natively generates synchronized, high-fidelity video and audio in a single pass from text, image, and video inputs. Key Features of LTX-2.5:
-
Synchronized Video and Audio: A dual-stream DiT with a 4096-wide video path, a 2048-wide audio path, and cross-modal attention joining them. No separate audio stage required.
-
Native Multishot Generation: Generates connected scenes in a single pass: multiple shots that hold character identity, environment, lighting, voice, and visual style across cuts (whereas previous versions produced a single continuous shot).
-
Diffusion Fidelity Rendering: Dynamically allocates compute by scene complexity and budget, rendering flawless detail where it matters and remaining efficient everywhere else.
-
Custom Gemma 4 12B Text Encoder: Holds complex prompts together (multiple characters, camera moves, lighting, actions) instead of dropping details across a longer sequence. The original bf16 transformer weighs in at 39 GB. These GGUF quantizations bring it down to a manageable 11-23 GB for consumer hardware.
GGUF Quantization VRAM Requirements
Below is a reference guide for the approximate VRAM required to load the 22 Billion parameter LTX-2.5-Distilled model at various GGUF quantization levels. Note: The sizes below refer specifically to the raw model weights loaded into memory. Generating video or high-resolution images will require an additional 4-8 GB of VRAM for the execution context, VAE, and text encoders depending on your ComfyUI configuration.
| Quantization Type | Bits per Weight (Avg) | Approx. Model Size (File / VRAM) | Recommended GPU VRAM (inc. Context) |
|---|---|---|---|
| Q2_K | 2.67 bits | ~6.8 GB | 10 GB - 12 GB GPU |
| Q3_K_S | 2.9 bits | ~7.8 GB | 12 GB - 16 GB GPU |
| Q4_K_M | 4.8 bits | ~ 11.5 GB | 16 GB - 20 GB GPU |
| Q5_K_M | 5.5 bits | ~ 14.2 GB | 20 GB - 24 GB GPU |
| Q6_K | 6.6 bits | ~ 16.8 GB | 24 GB GPU |
| Q8_0 | 8.5 bits | ~ 23.0 GB | 32 GB+ GPU (e.g. Mac / Dual GPU) |
Performance vs Quality
-
Q4_K_M: Best balance for typical consumer hardware (like a 16GB RTX 4080). Minor quality loss but runs very fast.
-
Q5_K_M: The sweet spot if you have a 24GB card (RTX 3090 / 4090). Noticeably better coherence than Q4 without hitting memory limits.
-
Q6_K: Very close to uncompressed BF16. Excellent choice if you have a powerful setup or Mac Studio.
-
Q8_0: Visually indistinguishable from the Master file, but requires significant memory.
Note: For image and video generation, the Q2 and Q3 model variants produce poor quality results and are generally unusable for final renders. They should only be used for basic testing and prototyping.
License & Disclaimer
Disclaimer: I am not the creator or owner of the LTX-2.5 model. This repository only provides quantized GGUF versions of the model for community use. All original licensing terms and usage restrictions carry over from the base model. The original model is released by Lightricks under the LTX-2.x Community License.
-
Under $10M annual revenue: Commercial and production use at no cost under the LTX-2.x Community License.
-
Over $10M annual revenue: Requires a Paid Commercial Use Agreement with Lightricks. Please refer to the original Lightricks/LTX-2.5 repository for full, binding license terms and to accept the gated model agreement.