v1.0 LoRA~7.6 GB16GB+ VRAMlora

ltx-2.3-22b-distilled-lora-384.safetensors

LTX 2.3 Distilled LoRA 384 (Official)

Official distilled LoRA rank-384 v1.0. Pair with dev model. Place in models/loras/.

Released 2026-03-04 · Source: Lightricks/LTX-2.3 (HuggingFace)Initial v1.0 distillation LoRA. Superseded by ltx-2.3-22b-distilled-lora-384-1.1.safetensors in the v1.1 release on 2026-04-13.

Download ltx-2.3-22b-distilled-lora-384.safetensors

Direct HuggingFace download. ~7.6 GB · Free.

Install path: ComfyUI/models/loras/ + ltx-2.3-22b-distilled-lora-384.safetensors

No 16GB GPU? Try ltx-2.3-22b-distilled-lora-384.safetensors online — free generation included

Skip the ~7.6 GB download and ComfyUI setup. Generate a 6-second video using this exact model in your browser, ~30 seconds.

Try this model online — free →

Technical details

ltx-2.3-22b-distilled-lora-384.safetensors is the official Lightricks distillation LoRA at rank 384 — the v1.0 release that shipped with the initial LTX 2.3 launch. Applying it on top of the dev model gives output equivalent to the distilled checkpoint: 8-step sampling with CFG=1 at distilled quality.

Rank 384 is comparatively high — high enough to capture most of the distillation behavior end-to-end without the parameter savings you'd see from a low-rank LoRA. The file is ~7.6 GB on disk, which is large for a LoRA but tiny next to the 42 GB base model it stacks on top of.

The rank-384 design lets Lightricks ship distillation as a swappable behavior change rather than a separate checkpoint — useful for training experiments where you want to compare dev + distill-LoRA against the standalone distilled model.

When to choose ltx-2.3-22b-distilled-lora-384.safetensors

For most users, switch to ltx-2.3-22b-distilled-lora-384-1.1.safetensors — the v1.1 release of the same LoRA. v1.1 has measurable improvements in fast-motion stability and prompt adherence over v1.0, with identical file structure.

This v1.0 file is worth keeping when you need to reproduce results from older workflow JSONs that pin the v1.0 filename, or when you're A/B testing v1.0 vs v1.1 to evaluate the upgrade.

For 16 GB inference where memory matters more than reproducibility, the Kijai dynamic LoRA (ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors) is ~3× smaller and loads faster.

Will this run on my GPU?

Minimum: 16GB VRAM.

GPUVRAMVerdict
RTX 3060 12GB12GBInsufficient VRAM
RTX 4060 Ti / 4070 (16GB)16GBTight fit
RTX 4070 Ti SUPER / 4080 (16GB)16GBTight fit
RTX 3090 (24GB)24GBComfortable
RTX 4090 (24GB)24GBComfortable
RTX 5090 / A6000 (32GB+)32GBComfortable

Recommendation: Previous v1.0 LoRA. Use v1.1 LoRA for latest quality.

How to use ltx-2.3-22b-distilled-lora-384.safetensors

  1. Download the file from HuggingFace.
  2. Place it in ComfyUI/models/loras/ inside your ComfyUI directory.
  3. Restart ComfyUI (or refresh the model list from the menu).
  4. Load a compatible workflow — see below.

Compatible official workflows:

Don't want to run this locally? Try ltx-2.3-22b-distilled-lora-384.safetensors online with a free generation — no GPU, no install, ~30 seconds per clip.

Common issues

LoRA loads but generation looks identical to base model — no distillation happening

Sampler is still configured for the dev model (30 steps, CFG > 1). Distillation LoRAs only show their effect with distilled sampling settings. Fix: Set sampler steps = 8 and CFG = 1 in the KSampler node. The LoRA's job is to make the dev model behave like the distilled model — it can't change the number of denoising steps for you.

Stacked this LoRA with the Kijai dynamic LoRA and got garbage output

Both LoRAs cover the same task — converting dev behavior to distilled behavior. Stacking them double-applies the distillation and overshoots. Fix: Use exactly one distillation LoRA at a time. If you want to compose with content/style LoRAs, apply the distillation LoRA at full strength and other LoRAs at their normal weights.

Workflow expects the v1.1 file, but only the v1.0 is in models/loras/

Workflow JSONs are filename-specific. v1.0 and v1.1 are not interchangeable by string. Fix: Either download ltx-2.3-22b-distilled-lora-384-1.1.safetensors (recommended), or edit the workflow's LoRA Loader node to point at the v1.0 file you have.

ComfyUI doesn't see the file after I downloaded it

Make sure the file is in ComfyUI/models/loras/ (not a subfolder). Restart ComfyUI fully — the menu refresh sometimes misses new files. Filename must match exactly: ltx-2.3-22b-distilled-lora-384.safetensors.

CUDA out of memory error when loading the model

ltx-2.3-22b-distilled-lora-384.safetensors needs ~16GB VRAM minimum. If you're hitting OOM: • Enable Sequential Offloading in ComfyUI settings • Lower the resolution (768×512 instead of 1280×704) — both dimensions must be divisible by 32 • Reduce frame count (65 frames instead of 161) — must be 8n+1 • Use a smaller variant — see Related models below.

How do I apply this LoRA in ComfyUI?

Load it in a 'LoraLoader' node and connect it after your model loader. Pair this LoRA with the dev base model (not the distilled one) for the right behavior. LoRA strength 1.0 is the trained value — start there.

Free newsletter

Get notified when LTX 2.3 Distilled LoRA 384 (Official) updates

Occasional updates on what's new in LTX 2.3 — new FP8 quants, LoRAs, IC-LoRA releases — with our hands-on verdict on whether they're worth re-downloading. No fixed cadence.

No spam. Sent occasionally when there's real news. Unsubscribe in one click.

Related models