I2V LoRA662 MB16GB+ VRAMlora

ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensors

LTX 2.3 Distilled 1.1 LoRA — fro90_ceil72 cond-safe (TenStrip)

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/.

Released 2026-05-01 · Source: TenStrip/LTX2.3_Distilled_Lora_1.1_Experiments (HuggingFace)Experimental community release in the v1.1 distillation LoRA experiments repo. Also mirrored at SulphurAI/Sulphur-2-base/distill_loras/.

Download ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensors

Direct HuggingFace download. 662 MB · Free.

Install path: ComfyUI/models/loras/ + ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensors

No 16GB GPU? Try ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensors online — free generation included

Skip the 662 MB 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 one of TenStrip's experimental distillation LoRAs from the LTX2.3_Distilled_Lora_1.1_Experiments repo. 'fro90_ceil72' names the training schedule's hyperparameters (Frobenius norm target 0.90, ceiling 72). The decisive marker is the 'condsafe' suffix.

In the 'cond-safe' variant, the LoRA matrices for cross-attention bridges, adaLN scale/shift tables, gate logits, and prompt scale-shift are zeroed out. Why this matters: those are the exact paths that carry input-image conditioning in I2V workflows. The standard dynamic-rank distillation LoRA modifies them along with the rest of the network — which speeds up T2V but can subtly break I2V because the image features get routed differently than the base model expects.

662 MB is small enough that you can keep this and the standard Kijai dynamic LoRA side-by-side and A/B test them per workflow. Place in ComfyUI/models/loras/ and select via the LoRA Loader node.

When to choose ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensors

Use this LoRA when your workflow is image-to-video, first-frame conditioning, or any pipeline where preserving input-image fidelity matters more than maximum T2V quality. The cond-safe design specifically protects the conditioning paths that I2V relies on.

For pure text-to-video, the standard Kijai dynamic LoRA (ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors) is more proven and produces stronger motion. The cond-safe variant trades some T2V quality for I2V reliability.

This is an experimental release — try it against the standard LoRA on your own prompts before committing a production workflow to it.

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 when running image-to-video or other input-conditioned workflows. Use the standard Kijai dynamic-rank LoRA for T2V instead.

How to use ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.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-1.1_fro90_ceil72_condsafe.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-lora-1.1_fro90_ceil72_condsafe.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:

  • ltx23\ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensors
  • distill_loras/ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensors
  • loras/ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.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-lora-1.1_fro90_ceil72_condsafe.safetensors — you have two fixes:

  1. Create the matching subdirectory inside ComfyUI/models/loras/ and place the file there. Example: if the workflow references ltx23\ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensors, create the corresponding subfolder under ComfyUI/models/loras/ and put ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.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-lora-1.1_fro90_ceil72_condsafe.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

I2V output looks blurry or loses the input image's structure

LoRA is paired with the distilled base instead of the dev base. Distillation LoRAs are designed to convert the dev model into distilled-quality output — applying them on top of an already-distilled model double-distills and degrades quality. Fix: Pair this LoRA with ltx-2.3-22b-dev-fp8.safetensors or the BF16 dev model, not the distilled FP8.

No noticeable speedup vs running the dev model raw

Distillation LoRAs only give the 8-step speedup if you also configure the sampler for distilled inference. Fix: Set sampler steps = 8 and CFG = 1 (matching the distilled model's inference profile). If you leave the sampler at the dev defaults (30 steps, CFG > 1), the LoRA does nothing useful.

ComfyUI errors with 'LoRA shape mismatch' when loading

Loading the LoRA on top of a transformer whose weights have a different layout — typically a quantized variant whose internal layer naming was rewritten. Fix: Apply this LoRA before quantization-aware loading, or use it with the BF16 dev base (ltx-2.3-22b-dev.safetensors) which has the canonical layer names.

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-1.1_fro90_ceil72_condsafe.safetensors.

CUDA out of memory error when loading the model

ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.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 — fro90_ceil72 cond-safe (TenStrip) 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