Official FP829.5 GB16GB+ VRAMfp8

ltx-2.3-22b-distilled-fp8.safetensors

LTX 2.3 Distilled FP8 (Official)

Official FP8 distilled model from Lightricks. 8 steps, CFG=1.

Released 2026-03-16 · Source: Lightricks/LTX-2.3-fp8 (HuggingFace)First-party FP8 distilled release. Same v1.0 weights as the non-FP8 distilled, quantized to FP8. Lightricks did not publish a v1.1 FP8 in this repo — use the Kijai mirror for v1.1.

Download ltx-2.3-22b-distilled-fp8.safetensors

Direct HuggingFace download. 29.5 GB · Free.

Install path: ComfyUI/models/checkpoints/ + ltx-2.3-22b-distilled-fp8.safetensors

No 16GB GPU? Try ltx-2.3-22b-distilled-fp8.safetensors online — free generation included

Skip the 29.5 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

Lightricks' first-party FP8 quantization of the v1.0 distilled model. Like the official dev FP8, this is a full checkpoint — transformer + VAE + text encoder integration glue — which is why the file is ~29.5 GB versus ~25 GB for a transformer-only quant.

The weights here are the v1.0 distilled checkpoint quantized to FP8. That means 8-step sampling with CFG=1, fast inference, and the v1.0 quality profile — not the v1.1 improvements that landed later in April 2026.

FP8 hardware (RTX 40-series Ada or newer) is required. On RTX 30-series the matmul falls back to a slow path; switch to the MXFP8 block-32 variant from Kijai if that's your GPU.

When to choose ltx-2.3-22b-distilled-fp8.safetensors

Use this when you need official Lightricks distilled weights at FP8 size — for example, building a reproducible reference pipeline or matching the exact outputs documented in Lightricks' v1.0 distilled release notes.

For most ComfyUI users, ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors (Kijai, v1.1) is a better default: newer weights, smaller file, and the filename every current workflow JSON references.

If you need v1.0 specifically (for reproducibility or comparison), this is the cleanest source — official, signed, in the same repo as the official dev FP8.

Will this run on my GPU?

Minimum: 16GB VRAM. Headroom up to: 24GB.

GPUVRAMVerdict
RTX 3060 12GB12GBInsufficient VRAM
RTX 4060 Ti / 4070 (16GB)16GBTight fit
RTX 4070 Ti SUPER / 4080 (16GB)16GBTight fit
RTX 3090 (24GB)24GBNo FP8 support
RTX 4090 (24GB)24GBComfortable
RTX 5090 / A6000 (32GB+)32GBComfortable

⚠ FP8 scaled matmul requires RTX 40-series or newer (Ada Lovelace architecture). RTX 30xx cannot run this format — use the MXFP8 block-32 or BF16 variant instead.

Recommendation: Official FP8 distilled from Lightricks. Alternative to Kijai's FP8.

How to use ltx-2.3-22b-distilled-fp8.safetensors

  1. Download the file from HuggingFace.
  2. Place it in ComfyUI/models/checkpoints/ 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-fp8.safetensors online with a free generation — no GPU, no install, ~30 seconds per clip.

Common issues

Output quality is lower than the Kijai v1.1 FP8 file at the same settings

This file is v1.0 weights, not v1.1. v1.1 is a meaningfully better release for fast motion and consistency. Fix: Switch to ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors if you don't need v1.0 specifically for reproducibility.

Workflow expects a 'transformer_only' file but downloaded this full checkpoint

Most community workflow JSONs load the transformer, VAE, and text encoder separately and expect Kijai-style transformer-only files. Fix: Either rewire the workflow to load the full checkpoint (single LoadCheckpoint node), or download ltx-2.3-22b-distilled_transformer_only_fp8_scaled.safetensors from Kijai for the v1.0 transformer-only equivalent.

Generation is slower than expected even on a 4090

Activations + Gemma BF16 text encoder + VAE add up. The 29.5 GB on-disk size grows substantially during inference. Fix: Use gemma_3_12B_it_fp4_mixed.safetensors as the text encoder. Cap resolution at 768p on 16 GB cards. On 24 GB you can run 1024p comfortably.

ComfyUI doesn't see the file after I downloaded it

Make sure the file is in ComfyUI/models/checkpoints/ (not a subfolder). Restart ComfyUI fully — the menu refresh sometimes misses new files. Filename must match exactly: ltx-2.3-22b-distilled-fp8.safetensors.

I get a CUDA error mentioning fp8 / scaled / matmul

FP8 scaled matmuls require an RTX 40-series GPU or newer (Ada Lovelace architecture). RTX 30-series and older cannot run FP8 weights at native precision. Use the BF16 variant instead, or the MXFP8 block-32 alternative.

CUDA out of memory error when loading the model

ltx-2.3-22b-distilled-fp8.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.

Free newsletter

Get notified when LTX 2.3 Distilled FP8 (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