LTX-2 22B
Frontierby Lightricks
LTX-2 (ltx-2.3-22b-dev) is Lightricks' DiT-based audio-video foundation model that generates synchronized video and audio in a single pass. ~22B parameters; a distilled variant runs at 8 steps with CFG=1 for fast generation. Width/height must be divisible by 32 and frame count must be divisible by 8 plus 1.
- ~22B DiT — joint synchronized audio + video generation in one model
- Distilled variant: 8 steps at CFG=1 for fast generation
- Width & height must be divisible by 32; frame count divisible by 8n+1
- Text-to-video, image-to-video, and video-to-video conditioning
Your hardware
Detecting...
Image Quality Benchmarks
Measured quality metrics for LTX-2 22B outputs.
How often humans prefer this model's output (0-100%)
Visual quality and composition rating (5-9 scale)
VRAM by Scenario
VRAM estimates at FP16 and FP8 precision. FP8 uses ~40% less memory with minimal quality loss. Grade shows how well each GPU handles the generation workload.
FP16 (full precision)
| Scenario | VRAM | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 · 25 frames | 61.8 GB | F | F | F | F |
| 768×512 · 25 frames | 63.9 GB | F | F | F | F |
| 768×512 · 100 frames | 70.2 GB | F | F | F | F |
| 1280×720 · 25 frames | 72.4 GB | F | F | F | F |
FP8 (quantized — ~40% less VRAM)
| Scenario | VRAM | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 · 25 frames | 33.8 GB | F | F | F | F |
| 768×512 · 25 frames | 35.9 GB | F | F | F | F |
| 768×512 · 100 frames | 42.2 GB | F | F | F | F |
| 1280×720 · 25 frames | 44.4 GB | F | F | F | F |
Run with Python
from diffusers import LTXConditionPipeline
import torch
pipe = LTXConditionPipeline.from_pretrained(
"Lightricks/LTX-2.3",
torch_dtype=torch.bfloat16
)
pipe.to("cuda")
frames = pipe(
prompt="your prompt here",
num_inference_steps=8,
guidance_scale=1.0,
num_frames=121,
).frames[0]
# Save frames or export as videoGet started
Setup instructions for running LTX-2 22B locally
1. Download the model
Get the checkpoint from HuggingFace
2. Place in:
ComfyUI/models/checkpoints/3. Launch ComfyUI
python main.pyMemory Breakdown
VRAM allocation for 25 frames at 768×512 on RTX 4090 24GB
Estimated Generation Time
25 frames at 768×512, 30 steps, FP16.
Available Formats & Downloads
Download LTX-2 22B in different precisions. Lower precision = less VRAM but slight quality loss.
| Format | Präzision | Größe | Anbieter | |
|---|---|---|---|---|
| Offizielle Gewichte | ||||
| safetensors | BF16 | 44.0 GB | official | Herunterladen |
| Offiziell quantisiert | ||||
| safetensorsEmpfohlenOffizielles FP8 | FP8 | 22.0 GB | official-fp8 | Herunterladen |
LoRA Ecosystem
LimitedEarly ecosystem for the LTX-2 generation; more LoRAs expected over time.
Related Workflows
You might also like
Frequently asked questions
FAQ — LTX-2 22B
How much VRAM does LTX-2 22B need for video?
LTX-2 22B (22B parameters) requires approximately 63.9 GB of VRAM at FP16 precision for generating 25 frames at 768×512. Video generation typically requires more VRAM than image generation due to temporal attention layers.
Can I run LTX-2 22B on RTX 4090?
LTX-2 22B exceeds the RTX 4090's 24 GB VRAM at FP16 for video generation. Consider reducing resolution, frame count, or using a GPU with more VRAM.
How long does it take to generate a video with LTX-2 22B?
On a reference GPU (RTX 4090 24GB), LTX-2 22B generates a 25-frame video at 768×512 in approximately ~3m 48s at FP16 with 30 inference steps. Faster GPUs with higher memory bandwidth will reduce generation time.
What resolution and frame count does LTX-2 22B support?
LTX-2 22B supports up to 1920×1088 resolution and 241 frames per generation at 30 FPS. Higher resolutions and frame counts require proportionally more VRAM.
About LTX-2 22B
See also