16GB LoRA2.74 GB16GB+ VRAMlora

ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors

LTX 2.3 Distilled 1.1 LoRA (Kijai)

Distilled LoRA v1.1 by Kijai. Use with the dev model for distilled-quality output on 16GB VRAM.

Released 2026-04-13 · Source: Kijai/LTX2.3_comfy (HuggingFace)v1.1 dynamic LoRA. Bumped average rank from 105 (v1.0) to 111 with refined Frobenius targets.

Download ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors

Direct HuggingFace download. 2.74 GB · Free.

Install path: ComfyUI/models/loras/ + ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors

No 16GB GPU? Try ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors online — free generation included

Skip the 2.74 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

This is Kijai's v1.1 dynamic-rank distillation LoRA. Applied on top of the dev model, it converts dev-style behavior (30 steps, CFG > 1) into distilled-style behavior (8 steps, CFG = 1) — giving you fast inference without downloading a separate 25 GB distilled checkpoint.

'Dynamic rank' means the LoRA matrices have variable rank per layer, averaging 111. Layers with higher information density (attention output projections, FFN gates) get more rank; less-critical layers get less. The 'fro09' in the filename references the Frobenius-norm target used during training (0.9 of the dev→distilled delta). The result is a 2.74 GB file that captures most of distilled quality at LoRA size.

The trick to using it is configuring the sampler correctly. Loading the LoRA without also switching to 8 steps + CFG=1 does nothing useful — the LoRA modifies the model's response to those specific sampler settings.

When to choose ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors

Use this LoRA when you want distilled-speed inference on the dev model — typically because you already have ltx-2.3-22b-dev-fp8.safetensors or the BF16 dev model and don't want to download a separate distilled checkpoint.

For straightforward distilled inference, ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors is more efficient — one file load, no LoRA composition overhead, identical quality.

Pick this LoRA over the r105 v1.0 dynamic LoRA (ltx-2.3-22b-distilled-lora-dynamic_fro09_avg_rank_105_bf16.safetensors) — v1.1 at rank 111 is the current best dynamic distillation LoRA.

For I2V workflows specifically, consider TenStrip's cond-safe LoRA instead — it zeroes the conditioning paths that standard distillation LoRAs disturb.

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: Pair with the dev FP8 model. Load as LoRA in ComfyUI models/loras/.

How to use ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.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-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors online with a free generation — no GPU, no install, ~30 seconds per clip.

ComfyUI says it can't find ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors?

Some published workflow JSONs reference this file under a custom subdirectory. If ComfyUI shows a "cannot find model" error and your workflow references one of these path-prefixed variants:

  • ltx2\ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors
  • ltx-2\ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors
  • loras/ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors

The prefix before the slash or backslash is a subdirectory the workflow author used. The actual file is the same ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors — you have two fixes:

  1. Create the matching subdirectory inside ComfyUI/models/loras/ and place the file there. Example: if the workflow references ltx2\ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors, create the corresponding subfolder under ComfyUI/models/loras/ and put ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors inside it.
  2. Or open the workflow JSON in a text editor and replace the prefixed string with just ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors. ComfyUI then resolves it directly from ComfyUI/models/loras/.

On Windows the separator is \, on macOS/Linux it is / — they refer to the same nested folder regardless of platform.

Common issues

LoRA loads but inference is still 30 steps and slow

Sampler still at dev defaults. The LoRA can't change step count for you — it only changes what the model produces given a specific sampler config. Fix: Set KSampler steps = 8 and CFG = 1. The LoRA is calibrated for these exact settings; deviating produces blurry or noisy output.

Output looks bad when stacked with a content/style LoRA

Stacking the distillation LoRA at full strength with another LoRA at full strength composes their deltas — the distillation effect can drown out the style LoRA, or vice versa. Fix: Keep the distillation LoRA at strength 1.0 (it needs full strength to do its job) and reduce content LoRAs to 0.6-0.8 strength. Or train the style LoRA against the distilled base directly to avoid stacking.

Applied on top of the distilled FP8 file and output is overcooked

Double-distillation. The distilled file is already distilled — this LoRA distills again, overshooting. Fix: Apply this LoRA only on dev-base files (ltx-2.3-22b-dev.safetensors, ltx-2.3-22b-dev-fp8.safetensors, ltx-2.3-22b-dev_transformer_only_*.safetensors). Never on a file with 'distilled' in the name.

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-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors.

CUDA out of memory error when loading the model

ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.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 1.1 LoRA (Kijai) 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