Run LTX 2.3 in ComfyUI — Match Your VRAM, Fix Missing Files, Generate the Workflow
Stuck on a missing taeltx2_3.safetensors, an out-of-memory error, or the wrong checkpoint for your GPU? Pick your VRAM, get the exact FP8 (16GB) or full-precision (32GB) files LTX 2.3 needs, and generate a ready-to-import ComfyUI workflow JSON in under 2 minutes. T2V, I2V, LoRA, and spatial upscaler all supported.
VRAM Adapter
Select your GPU VRAM to see recommended LTX 2.3 models:
Checkpoints for 16GB
ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensorsBest choice for 16GB VRAM. Latest v1.1 FP8 distilled. Requires RTX 40xx+ for fp8 matmuls.
ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensorsPair with the dev FP8 model. Load as LoRA in ComfyUI models/loras/.
ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensorsPair with the dev FP8 model when running image-to-video or other input-conditioned workflows. Use the standard Kijai dynamic-rank LoRA for T2V instead.
Required & optional components
ltx-2.3-spatial-upscaler-x1.5-1.0.safetensorsUse when x2 upscale is too aggressive. Gentler upscaling option.
ltx-2.3-temporal-upscaler-x2-1.0.safetensorsOptional. Use to double frame count (e.g. 65→129 frames) for smoother motion.
ltx-2.3-spatial-upscaler-x2-1.1.safetensorsOptional. Generate at a lower base resolution for speed, then run this x2 latent upscaler as a second stage for a sharper final result. Prefer v1.1 over v1.0. Tiny (~1 GB) and light on VRAM.
LTX 2.3 Model Downloads
All checkpoints for ComfyUI — FP8 (16GB), official distilled & dev (24–32GB), plus the required taeltx2_3.safetensors VAE.
LTX 2.3 ComfyUI Workflow JSON
Official workflow JSON files from Lightricks — T2V, I2V, two-stage upscaler, ICLoRA motion tracking. Drag into ComfyUI to load.
ComfyUI Workflow Configurator
Answer a few questions — we'll pick the right workflow, model files, and node settings for your GPU.
⚠ RTX 30xx? The selected FP8 model needs RTX 40xx+ hardware. Use the int8_convrot variant (fastest on Ampere INT8 cores), or mxfp8_block32 / BF16 instead.
Models for your GPU
Transformer · 768x512
ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensorsText encoder (Gemma 3 12B)
gemma_3_12B_it_fp4_mixed.safetensorsFP8 transformer (~25 GB) + FP4 Gemma (9.5 GB)
Recommended workflow
T2V / I2V Single Stage Distilled
Standard I2V workflow — attach your image to LTXVImgToVideoConditioning. Generates video + synchronized audio in one pass.
!IMPORTANT: After loading the JSON, update LoraLoaderModelOnly → ltx-2.3-22b-distilled-lora-384-1.1.safetensors. The official JSON ships v1.0 refs — v1.0 LoRA on v1.1 base causes stiff motion and color drift.
Required files
- ✓
ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors - ✓
taeltx2_3.safetensors - ✓
LTX23_audio_vae_bf16.safetensors - ✓
gemma_3_12B_it_fp4_mixed.safetensors - ✓
ltx-2.3-22b-distilled-lora-384-1.1.safetensors
Node settings after loading the workflow
ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensorsLTX23_audio_vae_bf16.safetensorsPlace in models/checkpoints/
gemma_3_12B_it_fp4_mixed.safetensorsyour-image.pngDrag your source image here
768 / 512Both must be divisible by 32
121Must be 8n+1 (65, 97, 121…)
241Keep at 1.0 — raising CFG doubles VRAM with no quality benefit
euler_ancestral_cfg_ppVerified from official JSON
ltx-2.3-22b-distilled-lora-384-1.1.safetensorsstrength 0.5 — must be v1.1 (official JSON ships v1.0 by mistake)
LTX Insider
Occasional updates when notable LTX 2.3 models drop — Kijai FP8 quants, Lightricks releases, IC-LoRAs. No fixed cadence, no spam. Unsubscribe in one click.
No spam. Sent occasionally when there's real news. Unsubscribe in one click.
Official Resources
LTX 2.3 on HuggingFace →
Official model weights by Lightricks
ComfyUI-LTXVideo →
Official ComfyUI nodes & example workflows
Kijai FP8 Models →
FP8 quantized variants for 16GB VRAM
LTX-Video GitHub →
Official LTX-Video model repository
ComfyUI →
Node-based UI for running diffusion models
ComfyUI Manager →
Install LTXVideo nodes via Manager