LTX-2 VRAM Requirements
LTX-2 needs about 19.2 GB of VRAM with FP8 weights at 768×512 and 121 frames, loading one model at a time. In FP16 it needs 36.9 GB (61.4 GB with everything loaded), and GGUF Q4 brings it to 12.6 GB. The smallest consumer GPU that runs it is the RTX 4060 Ti 16GB with Q4 weights, or the Radeon RX 7900 XT 20GB in FP8.
19.2 GB peak VRAM
LTX-2 at 768×512, 121 frames with FP8 weights, loading one model at a time (ComfyUI default on smaller cards). Keeping everything loaded needs 32.4 GB.
- Diffusion model 17.7 GB
- Text encoders 11.4 GB
- VAE 0.9 GB
- Peak activations 1.6 GB
- Overhead 0.8 GB
GPU fit
- RTX 4060 8GB39% on GPU
- RTX 3060 12GB59% on GPU
- RTX 4060 Ti 16GB78% on GPU
- RTX 4090 24GBFitswith swapping
- RTX 5090 32GBFitswith swapping
- RTX 6000 Ada 48GBFitsall loaded
- RTX PRO 6000 Blackwell 96GBFitsall loaded
- Apple M4 Max 128GBFitsall loaded
LTX-2 VRAM by precision
At 768×512, 121 frames, tiled VAE. Text encoders in FP8 except for the FP16 row.
| Weights | Diffusion model | Peak (swapping) | All loaded | Smallest tier |
|---|---|---|---|---|
| FP16 / BF16 | 35.4 GB | 36.9 GB | 61.4 GB | 48 GB |
| FP8 | 17.7 GB | 19.2 GB | 32.4 GB | 24 GB |
| GGUF Q8_0 | 18.8 GB | 20.4 GB | 33.5 GB | 24 GB |
| GGUF Q5_K_S | 12.2 GB | 13.8 GB | 26.9 GB | 16 GB |
| GGUF Q4_K_S / NF4 | 10.1 GB | 12.6 GB | 24.8 GB | 16 GB |
VRAM by resolution
FP8 weights, with model swapping.
| Resolution | Latent tokens | Activations | Peak VRAM |
|---|---|---|---|
| 832×480, 121 frames | 6,240 | 0.8 GB | 19.3 GB |
| 1280×720, 121 frames | 14,720 | 1.8 GB | 20.3 GB |
About LTX-2
A 19B model that generates video and synchronized audio together. Its VAE compresses heavily (32× in space, 8× in time), so the transformer sees far fewer tokens than Wan or HunyuanVideo at the same resolution. The Gemma 3 12B text encoder is the other large piece; FP8 and NVFP4 checkpoints are available.
| Component | Size |
|---|---|
| Diffusion model | 19B parameters (Audio-video diffusion transformer) |
| Text encoders | Gemma 3 12B (12.2B) |
| VAE | 0.5B parameters |
| Classifier-free guidance | Yes (2 passes per step) |
| Released | 2026-01 |
| License | LTX-2 Community License |
LTX-2 VRAM: common questions
Can LTX-2 run on a 16GB or 24GB GPU?
On 24 GB, yes with FP8 weights and model swapping. On 16 GB use GGUF Q4–Q5 weights, a quantized text encoder and tiled VAE decoding.
How much VRAM does LTX-2 need?
At its default 768×512 and 121 frames: about 36.9 GB in FP16 with model swapping (61.4 GB with everything loaded), 19.2 GB in FP8 and 12.6 GB with GGUF Q4 weights.