ltx-2.3-22b-dev-fp8.safetensors
LTX 2.3 Dev FP8 (Official)
Official FP8 dev model from Lightricks. 29.1GB, runs on 16GB VRAM.
Released 2026-03-16 · Source: Lightricks/LTX-2.3-fp8 (HuggingFace) — Official first-party FP8 release from Lightricks, posted shortly after the initial LTX 2.3 launch.
Download ltx-2.3-22b-dev-fp8.safetensors
Direct HuggingFace download. 29.1 GB · Free.
No 16GB GPU? Try ltx-2.3-22b-dev-fp8.safetensors online — free generation included
Skip the 29.1 GB download and ComfyUI setup. Generate a 6-second video using this exact model in your browser, ~30 seconds.
Technical details
This is Lightricks' first-party FP8 quantization of the LTX 2.3 dev base model. Unlike Kijai's transformer-only variants, the official file is a complete checkpoint — it includes the full set of submodules needed to run inference, which is why it weighs ~29 GB compared to ~25 GB for transformer-only quants.
'Dev' means this is the non-distilled base model. It runs at full 30-step sampling with CFG > 1, which is slower than the 8-step distilled path but more flexible: it accepts LoRAs cleanly, responds to negative prompts, and is the right base for fine-tuning experiments.
FP8 hardware is required (RTX 40-series Ada or newer, H100, Blackwell). On Ampere or older you'll either crash or fall back to a slow emulation path — switch to the BF16 dev file in that case.
When to choose ltx-2.3-22b-dev-fp8.safetensors
Use this when you specifically want official Lightricks FP8 weights — for example, you're building a service that needs first-party-signed weights, or you're matching the exact reference output documented in Lightricks' release notes.
For most users, ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors is the better default — it's smaller, newer (v1.1), faster (8 steps), and from the community-trusted Kijai mirror that all current ComfyUI workflows reference. The official dev FP8 is what you reach for when you need LoRA support at FP8 size and the distilled path doesn't fit your use case.
Will this run on my GPU?
Minimum: 16GB VRAM. Headroom up to: 24GB.
⚠ 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 from Lightricks. Alternative to Kijai's FP8 dev model.
How to use ltx-2.3-22b-dev-fp8.safetensors
- Download the file from HuggingFace.
- Place it in ComfyUI/models/checkpoints/ inside your ComfyUI directory.
- Restart ComfyUI (or refresh the model list from the menu).
- Load a compatible workflow — see below.
Don't want to run this locally? Try ltx-2.3-22b-dev-fp8.safetensors online with a free generation — no GPU, no install, ~30 seconds per clip.
Common issues
Workflow expects a 'transformer_only' filename but I downloaded the full checkpoint▼
Most published ComfyUI workflows are written against Kijai's transformer_only files, not Lightricks' full checkpoints. Fix: Either rewire the workflow's LoadCheckpoint to point at this file (instead of separate transformer + VAE + text encoder loads), or switch to ltx-2.3-22b-dev_transformer_only_fp8_scaled.safetensors from Kijai.
Generation looks correct but is slower than expected on a 4090▼
This is the dev model, not distilled. It runs at 30 steps by default vs the distilled model's 8 steps. Fix: Either accept the speed for higher fidelity, or switch to ltx-2.3-22b-distilled-fp8.safetensors (also official, distilled) and use 8 steps with CFG=1.
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-dev-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-dev-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.
Get notified when LTX 2.3 Dev 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
ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors
ltx-2.3-22b-distilled-1.1_transformer_only_mxfp8_block32.safetensors
ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled_v3.safetensors
ltx-2.3-22b-dev_transformer_only_fp8_input_scaled.safetensors
ltx-2.3-22b-dev_transformer_only_fp8_scaled.safetensors
ltx-2.3-22b-dev_transformer_only_mxfp8_block32.safetensors