ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.safetensors
LTX 2.3 Distilled FP8 v1 (Kijai)
FP8 distilled v1 by Kijai. Earliest FP8 release, superseded by v3.
Released 2026-03 · Source: Kijai/LTX2.3_comfy (HuggingFace) — First FP8 input-scaled release for distilled. Superseded by v2, then v3 — all v1.0 weights — then by v1.1 line on 2026-04-13.
Download ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.safetensors
Direct HuggingFace download. ~25 GB · Free.
No 16GB GPU? Try ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.safetensors online — free generation included
Skip the ~25 GB download and ComfyUI setup. Generate a 6-second video using this exact model in your browser, ~30 seconds.
Technical details
The first FP8 input-scaled distilled quantization Kijai shipped — no version suffix in the filename because it predated the v2/v3 iterations. Same v1.0 distilled weights as v3, but with the initial calibration that v2 and v3 later refined.
'input_scaled' applies FP8 scaling on activations entering each linear layer rather than on the weights themselves. This was Kijai's first FP8 attempt; calibration issues in v1 caused subtle quality regressions on certain prompt distributions, which v3 fixed by re-running calibration on a larger prompt set.
File size and inference profile match v3. The only meaningful difference is quality on edge-case prompts (very short prompts, prompts with unusual tokens, mixed-language prompts).
When to choose ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.safetensors
Almost never. This is the v1.0-era v1 calibration — older weights AND older calibration. It exists in GSC traffic because workflow JSONs from early ComfyUI ports still reference this exact filename.
If a workflow forces you to keep this file, fine — but for anything you control, switch to ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors (v1.1 weights, current calibration, weight-scaled instead of input-scaled, runs on the same RTX 40xx+ hardware).
The single use case for keeping v1: precisely reproducing output from a very old workflow that locks in the v1 calibration's quirks.
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: Previous version. Use FP8 v3 or v1.1 FP8 for better quality.
How to use ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.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.
Compatible official workflows:
- LTX-2.3_T2V_I2V_Single_Stage_Distilled_Full.json— T2V / I2V Single Stage Distilled
- LTX-2.3_T2V_I2V_Two_Stage_Distilled.json— T2V / I2V Two Stage Distilled
- LTX-2.3_ICLoRA_Union_Control_Distilled.json— ICLoRA Union Control Distilled
- LTX-2.3_ICLoRA_Motion_Track_Distilled.json— ICLoRA Motion Track Distilled
- LTX-2.3_ICLoRA_HDR_Distilled.json— ICLoRA HDR Distilled
Don't want to run this locally? Try ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.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_transformer_only_fp8_input_scaled.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:
- ltx\ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.safetensors
- diffusion_models/ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.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_transformer_only_fp8_input_scaled.safetensors — you have two fixes:
- Create the matching subdirectory inside ComfyUI/models/checkpoints/ and place the file there. Example: if the workflow references ltx\ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.safetensors, create the corresponding subfolder under ComfyUI/models/checkpoints/ and put ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.safetensors inside it.
- Or open the workflow JSON in a text editor and replace the prefixed string with just ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled.safetensors. ComfyUI then resolves it directly from ComfyUI/models/checkpoints/.
On Windows the separator is \, on macOS/Linux it is / — they refer to the same nested folder regardless of platform.
Common issues
Output quality is noticeably worse than v3 or v1.1▼
v1 calibration had under-coverage on certain prompt types. v2 partially fixed, v3 fully fixed. Fix: Switch to v3 (ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled_v3.safetensors) or v1.1 (ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors). Both are strict upgrades.
Workflow references this exact filename — can I just rename v3 to match?▼
ComfyUI matches by exact filename string. Fix: Yes, but you'll lose the version tag in your install. Better: edit the workflow JSON's loader string to point at the v3 or v1.1 file you actually want. Saves you from confusion later when you can't tell which version is which.
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_transformer_only_fp8_input_scaled.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_transformer_only_fp8_input_scaled.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 Distilled FP8 v1 (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
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