LTX 2.3 Model Downloads
Download all LTX 2.3 model files for ComfyUI with direct HuggingFace links — organized by GPU VRAM. Start with taeltx2_3.safetensors (VAE, required for all setups), then choose a checkpoint: FP8 quantized for 16GB VRAM or official full precision for 32GB+.
No GPU? Try LTX 2.3 online — free generation included
Skip the 16GB+ VRAM card and ComfyUI setup. Image-to-video, no install, ~30 seconds per clip.
Quick Start: 3 Steps to Get Running
- 1.Download taeltx2_3.safetensors (VAE) — required for all workflows
- 2.Choose your checkpoint: FP8 v3 (16GB VRAM) or Official Distilled (32GB VRAM)
- 3.Place files in ComfyUI folders: VAE →
models/vae/, Checkpoint →models/checkpoints/
16GB VRAM — FP8 / MXFP8 Quantized (RTX 40xx+)
FP8 scaled requires RTX 40-series or newer. MXFP8 block-32 is an alternative format for compatible GPUs. NVFP4 (21.7 GB) is the official Blackwell / RTX 50xx path. Use v1.1 FP8 Distilled for fastest generation; use Dev FP8 + LoRA v1.1 if applying LoRA weights.
LTX 2.3 Dev FP8 (Official)
Official FP8 from Lightricks. Alternative to Kijai's FP8 dev model.
Official FP8 dev model from Lightricks. 29.1GB, runs on 16GB VRAM.
LTX 2.3 Distilled FP8 (Official)
Official FP8 distilled from Lightricks. Alternative to Kijai's FP8.
Official FP8 distilled model from Lightricks. 8 steps, CFG=1.
LTX 2.3 Distilled 1.1 FP8 (Kijai)
Best choice for 16GB VRAM. Latest v1.1 FP8 distilled. Requires RTX 40xx+ for fp8 matmuls.
FP8 quantized v1.1 distilled by Kijai. Best for 16GB VRAM. 8 steps, CFG=1.
LTX 2.3 Distilled 1.1 MXFP8 (Kijai)
RTX 30xx workaround — use this when your GPU lacks RTX 40xx-style FP8 matmul support. Same VRAM as fp8_scaled.
MXFP8 block-32 quantized distilled 1.1 by Kijai. Use on RTX 30xx GPUs that cannot run standard FP8 scaled matmul.
LTX 2.3 Dev FP8 (Kijai)
Use this (not distilled) if you want to apply LoRA weights on 16GB VRAM.
FP8 quantized dev model by Kijai. Runs on 16GB VRAM. Supports LoRA. Place in models/checkpoints/.
LTX 2.3 Dev FP8 Scaled (Kijai)
Alternative FP8 quantization for dev. Use when fp8_input_scaled has compatibility issues.
FP8 scaled dev model by Kijai. 16GB VRAM. Supports LoRA. Place in models/checkpoints/.
LTX 2.3 Dev MXFP8 (Kijai)
RTX 30xx workaround for dev/quality path — use when standard FP8 matmul is unsupported. Supports LoRA.
MXFP8 block-32 dev model by Kijai. 16GB VRAM. Supports LoRA. Use on RTX 30xx GPUs that cannot run standard FP8 scaled. Place in models/checkpoints/.
LTX 2.3 Dev NVFP4 (Official)
Best size/quality tradeoff on RTX 50xx. On older GPUs it falls back to slow paths — prefer FP8 there.
NVFP4 quantized dev model. Native nvfp4 matmul on Blackwell GPUs (RTX 50xx). 21.7GB.
LTX 2.3 Distilled LoRA 384 v1.1 (Official)
Latest official LoRA. Pair with dev model for distilled-quality output.
Official distilled LoRA rank-384 v1.1. Use with dev model. Place in models/loras/.
LTX 2.3 Distilled 1.1 LoRA (Kijai)
Pair with the dev FP8 model. Load as LoRA in ComfyUI models/loras/.
Distilled LoRA v1.1 by Kijai. Use with the dev model for distilled-quality output on 16GB VRAM.
LTX 2.3 Distilled 1.1 LoRA — fro90_ceil72 cond-safe (TenStrip)
Pair 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.
Experimental community distilled LoRA v1.1 by TenStrip. The 'cond-safe' variant zeroes cross-attention bridges, adaln/scale-shift tables, gate logits, and prompt scale-shift — making it better suited for I2V and input-conditioned workflows than the standard dynamic LoRA. Place in models/loras/.
LTX 2.3 Distilled LoRA Dynamic r105 (Kijai)
v1.0 dynamic LoRA. Use v1.1 dynamic LoRA (rank 111) for latest quality.
Dynamic rank LoRA (avg rank 105) by Kijai. Pair with dev model. Place in models/loras/.
24GB VRAM — Official + Sequential Offloading
Enable sequential offloading in ComfyUI settings (Model Offload or Sequential). Uses latest v1.1 official weights.
LTX 2.3 Distilled 1.1 (bf16, 24GB)
Enable sequential offloading in ComfyUI settings. Uses latest v1.1 official weights.
Official v1.1 distilled model runnable on 24GB with sequential offloading enabled in ComfyUI.
32GB VRAM — Official Full Precision (BF16)
Official Lightricks checkpoints at full bf16 precision. v1.1 Distilled recommended for most use cases (8 steps, CFG=1). BF16 transformer-only variants from Kijai are also available.
LTX 2.3 Distilled 1.1
For 32GB VRAM: enable Sequential Offloading in ComfyUI settings (file is 46GB). For fastest inference, use the FP8 distilled variant instead.
Official v1.1 distilled model. 8 steps, CFG=1. Latest release from Lightricks. Requires sequential offloading on 32GB — file is 46GB.
LTX 2.3 Distilled 1.1 BF16 (Kijai)
Use when FP8 matmul (RTX 40xx+) is unavailable. 44GB — requires 48GB VRAM or sequential offloading on 32GB.
BF16 distilled 1.1 transformer-only by Kijai. For RTX 30xx or GPUs without FP8 support. Requires 48GB+ or sequential offloading on 32GB.
LTX 2.3 Dev
Best for LoRA training and fine-tuning. 42GB — enable Sequential Offloading on 32GB cards.
Full BF16 dev model. Flexible and trainable. 42GB — requires 48GB VRAM or sequential offloading on 32GB.
LTX 2.3 Dev BF16 (Kijai)
Use when training LoRA with full BF16 precision. 44GB — enable Sequential Offloading on 32GB cards.
BF16 dev transformer-only by Kijai. 44GB — requires 48GB VRAM or sequential offloading on 32GB. Place in models/checkpoints/.
LTX 2.3 Distilled LoRA 384 v1.1 (Official)
Latest official LoRA. Pair with dev model for distilled-quality output.
Official distilled LoRA rank-384 v1.1. Use with dev model. Place in models/loras/.
LTX 2.3 Distilled LoRA 384 (Official)
Previous v1.0 LoRA. Use v1.1 LoRA for latest quality.
Official distilled LoRA rank-384 v1.0. Pair with dev model. Place in models/loras/.
IC-LoRA Family — Control, HDR, LipDub (Official Lightricks)
IC-LoRAs (In-Context LoRAs) attach to the dev model for structural control (pose/depth/edges), motion trajectories, 16-bit HDR generation, and lip-dub. Use the official IC-LoRA ComfyUI workflow with LTXICLoRALoaderModelOnly + LTXAddVideoICLoRAGuide at Reference Downscale Factor 0.5. Place in models/loras/. — See Guide step 7 for the full how-to.
IC-LoRA Union Control
Use the official IC-LoRA ComfyUI workflow with LTXICLoRALoaderModelOnly + LTXAddVideoICLoRAGuide (Reference Downscale Factor 0.5).
Unified Canny + Depth control IC-LoRA. One LoRA covering edge and depth conditioning. Place in models/loras/.
IC-LoRA Motion Track Control
For V2V motion transfer. Pair with the official IC-LoRA workflow at Reference Downscale Factor 0.5.
Motion-track IC-LoRA. Drive motion with sparse point trajectories (SpatialTrackerV2 or hand-drawn). Place in models/loras/.
IC-LoRA HDR
Required: also download the HDR scene-emb file. Enables HDR text/image-to-video and SDR-to-HDR conversion.
16-bit HDR generation IC-LoRA. SDR→HDR via LogC3 transform. Place in models/loras/.
IC-LoRA HDR Scene Embeddings
Required when using the HDR IC-LoRA.
Companion scene-embeddings file for the HDR IC-LoRA. Place in models/loras/.
IC-LoRA LipDub
Use for video lip-sync / dubbing. Pair with the audio VAE.
Lip-dubbing IC-LoRA based on JustDubIt research. Joint audio-visual diffusion for dubbing. Place in models/loras/.
Text Encoders — Gemma 3 12B IT (Required)
Every LTX 2.3 ComfyUI workflow needs a Gemma 3 12B text encoder. Use FP4 mixed (9.5 GB) on 16/24 GB cards; the full BF16 file is for 32 GB+. Workflows reference the BF16 file as 'comfy_gemma_3_12B_it.safetensors' — rename after download. Place in models/text_encoders/. — See Guide step 5.
Gemma 3 12B IT FP4 Mixed (Text Encoder)
Use this on 16/24 GB VRAM cards. Full BF16 Gemma OOMs alongside the transformer.
FP4-mixed Gemma 3 12B IT text encoder (~90% FP4 layers). Required for 16-24 GB ComfyUI workflows. Place in models/text_encoders/.
Gemma 3 12B IT FP8 Scaled (Text Encoder)
Alternative for 16-24 GB cards. Slightly higher quality than FP4 at 3.7 GB more.
FP8-scaled Gemma 3 12B IT text encoder. Alternative to FP4 with marginally higher precision. Place in models/text_encoders/.
Gemma 3 12B IT BF16 (Text Encoder)
Rename to comfy_gemma_3_12B_it.safetensors after downloading. Use on 32 GB+ cards for highest text-encoder quality.
Full BF16 Gemma 3 12B IT text encoder (HF filename gemma_3_12B_it.safetensors — workflows reference it as comfy_gemma_3_12B_it.safetensors, rename after download). Place in models/text_encoders/.
LTX 2.3 Text Projection (Kijai)
Required for workflows using separate text encoder components.
Text projection BF16 component by Kijai. Place in models/text_encoders/.
Previous Versions — v1.0 Models
v1.0 variants superseded by v1.1. Listed for reference or compatibility with existing workflows.
LTX 2.3 Distilled
Previous version. Use v1.1 Distilled for latest quality.
v1.0 distilled version. 8 steps, CFG=1. Superseded by v1.1.
LTX 2.3 Distilled FP8 v3 (Kijai)
Previous version. Use v1.1 FP8 for latest quality.
FP8 distilled v3 by Kijai. Previous version, superseded by v1.1 FP8.
LTX 2.3 Distilled FP8 v1 (Kijai)
Previous version. Use FP8 v3 or v1.1 FP8 for better quality.
FP8 distilled v1 by Kijai. Earliest FP8 release, superseded by v3.
LTX 2.3 Distilled FP8 v2 (Kijai)
Previous version. Use FP8 v3 or v1.1 FP8 for better quality.
FP8 distilled v2 by Kijai. Superseded by v3.
LTX 2.3 Distilled FP8 Scaled (Kijai)
Alternative FP8 quantization. Use v1.1 FP8 for latest quality.
FP8 scaled distilled by Kijai. Alternative FP8 quantization method.
LTX 2.3 Distilled MXFP8 (Kijai)
Previous v1.0 MXFP8 (RTX 30xx). Use v1.1 MXFP8 for latest quality.
MXFP8 block-32 distilled v1.0 by Kijai. RTX 30xx workaround — use when standard FP8 scaled is unsupported.
LTX 2.3 Distilled BF16 (Kijai)
Previous v1.0 BF16 distilled. Use v1.1 BF16 for latest quality. Enable Sequential Offloading on 32GB.
BF16 distilled v1.0 transformer-only by Kijai. 44GB — requires 48GB VRAM or sequential offloading on 32GB.
LTX 2.3 Dev FP8 Input Scaled (dash filename, Kijai)
Prefer the dotted-filename version unless a specific workflow references this exact name.
FP8 input-scaled dev model under the alternate `ltx-2-3-…` (dash) filename. Same weights, different name.
Required & Optional Components
taeltx2_3.safetensors (VAE) is required for all setups. Audio VAE enables audio-conditioned workflows. Upscalers are optional — place in models/latent_upscale_models/.
LTX 2.3 VAE
Required for all setups. Download this first regardless of your VRAM.
VAE by Kijai. Required for all ComfyUI workflows. Place in models/vae/.
LTX 2.3 Audio VAE (Kijai)
Required for audio-conditioned workflows. Unlocks new audio-to-video capability.
Audio VAE for audio-conditioned video generation. Place in models/vae/. Enables audio-to-video workflows.
LTX 2.3 Video VAE (Kijai)
Alternative VAE option. Use in workflows that require the separate BF16 VAE component.
Standalone video VAE BF16 by Kijai. Alternative to taeltx2_3. Place in models/vae/.
Spatial Upscaler x2 v1.1
Updated upscaler. Prefer over v1.0 for better upscaling quality.
Updated spatial upscaler x2 v1.1. Place in models/latent_upscale_models/.
Spatial Upscaler x2
Optional. Use in a two-stage pipeline to upscale output resolution after generation.
Spatial upscaler x2 for two-stage pipelines. Place in models/latent_upscale_models/.
Spatial Upscaler x1.5
Use when x2 upscale is too aggressive. Gentler upscaling option.
Spatial upscaler x1.5 for two-stage pipelines. Place in models/latent_upscale_models/.
Temporal Upscaler x2
Optional. Use to double frame count (e.g. 65→129 frames) for smoother motion.
Temporal upscaler x2 for frame interpolation. Place in models/latent_upscale_models/.
How to Choose the Right Model
Distilled vs Dev
Distilled — 8 steps, CFG=1. Recommended for most users. Fastest generation, high quality. Cannot be fine-tuned.
Dev — Full model. Use only if you need LoRA training or fine-tuning. Slower, more flexible.
FP8 vs MXFP8 vs BF16
FP8 scaled (Kijai) — Standard 8-bit float. Runs on 16GB. Requires RTX 40xx+. Slight quality trade-off.
MXFP8 block-32 (Kijai) — Alternative FP8 format. Try if standard FP8 causes errors on your GPU.
BF16 (Official / Kijai) — Full precision. Best quality. Needs 32GB+ VRAM.
Quick decision
- → 16GB, RTX 40xx+: Distilled 1.1 FP8 scaled
- → 16GB, FP8 issues: Distilled 1.1 MXFP8 block-32
- → 16GB + LoRA: Dev FP8 + LoRA 1.1
- → 24GB: Official Distilled 1.1 + offloading
- → 32GB+: Official Distilled 1.1 (recommended)
Always required
taeltx2_3.safetensors (VAE) — place in models/vae/. Every workflow needs this regardless of which checkpoint you use. For audio-to-video workflows, also download LTX23_audio_vae_bf16.safetensors.