ltx-2.3-22b-distilled-1.1_transformer_only_int8_convrot.safetensors
LTX 2.3 Distilled 1.1 INT8 convrot (Kijai)
INT8 convrot quantized distilled 1.1 by Kijai. Runs on RTX 30xx (Ampere INT8 tensor cores) — the smallest distilled transformer that keeps near-FP8 quality without needing FP8 matmul.
Released 2026-06 · Source: Kijai/LTX2.3_comfy (HuggingFace) — Added to bring hardware-accelerated low-VRAM inference to RTX 20xx/30xx, which can't use FP8 matmul and only get BF16-fallback speed from MXFP8.
Download ltx-2.3-22b-distilled-1.1_transformer_only_int8_convrot.safetensors
Direct HuggingFace download. ~21.5 GB · Free.
No 16GB GPU? Try ltx-2.3-22b-distilled-1.1_transformer_only_int8_convrot.safetensors online — free generation included
Skip the ~21.5 GB download and ComfyUI setup. Generate a 6-second video using this exact model in your browser, ~30 seconds.
Technical details
INT8 convrot is a rotation-based INT8 quantization: the weights are multiplied by a learned/computed rotation ('convrot') before being cast to 8-bit integers, which spreads out the activation outliers that normally wreck naive INT8. The result is INT8 memory footprint with quality much closer to FP8/BF16 than plain per-tensor INT8 would give.
The practical win over ltx-2.3-22b-distilled-1.1_transformer_only_mxfp8_block32.safetensors is hardware reach: INT8 matmul tensor cores exist on every NVIDIA card since Turing (RTX 20xx) and Ampere (RTX 30xx), so you get real compute acceleration — not the BF16-fallback path MXFP8 takes on those GPUs. File size is also a bit smaller (~21.5 GB vs ~25 GB).
'transformer_only' means DiT weights only. Pair it with taeltx2_3.safetensors (VAE) and a Gemma 3 12B text encoder (FP4 mixed on 16 GB, FP8 scaled or BF16 on more). Distilled inference settings: 8 steps, CFG=1.
When to choose ltx-2.3-22b-distilled-1.1_transformer_only_int8_convrot.safetensors
Pick INT8 convrot on RTX 20-series or 30-series (2080 Ti, 3060 12GB, 3080, 3090) when you want the distilled path and want actual matmul speedup, not just a VRAM fit. Ampere lacks FP8 matmul, and MXFP8 falls back to BF16 compute there — INT8 convrot uses the INT8 tensor cores those cards do have.
On RTX 40-series (Ada) or 50-series (Blackwell), the standard ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors is still the default — native FP8 is marginally higher quality and equally fast. Reach for INT8 convrot on those cards only if you're memory-constrained and want the smaller file.
Requires a recent ComfyUI + KJNodes that understands the convrot layout — older builds will error on load.
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: Best low-VRAM distilled option for RTX 30xx. Smaller than MXFP8 (~21.5 GB) and runs on INT8 tensor cores every Ampere+ card has.
How to use ltx-2.3-22b-distilled-1.1_transformer_only_int8_convrot.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-1.1_transformer_only_int8_convrot.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_transformer_only_int8_convrot.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-1.1_transformer_only_int8_convrot.safetensors
- diffusion_models/ltx-2.3-22b-distilled-1.1_transformer_only_int8_convrot.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_transformer_only_int8_convrot.safetensors — you have two fixes:
- Create the matching subdirectory inside ComfyUI/models/checkpoints/ and place the file there. Example: if the workflow references ltx23\ltx-2.3-22b-distilled-1.1_transformer_only_int8_convrot.safetensors, create the corresponding subfolder under ComfyUI/models/checkpoints/ and put ltx-2.3-22b-distilled-1.1_transformer_only_int8_convrot.safetensors inside it.
- Or open the workflow JSON in a text editor and replace the prefixed string with just ltx-2.3-22b-distilled-1.1_transformer_only_int8_convrot.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
ComfyUI errors on load: unknown quantization / unexpected keys▼
The convrot rotation tables are stored as extra tensors the loader must apply before matmul. Older ComfyUI or KJNodes builds don't recognize them. Fix: Update ComfyUI and KJNodes to a build from 2026-06 or later. Load with the Kijai LTXVideoModelLoader (KJNodes), not a custom loader that assumes plain INT8.
Output has color banding or blocky artifacts vs the FP8 file▼
A LoRA or loader is applying deltas in the wrong dtype, bypassing the rotation. Fix: Confirm the base runs clean with no LoRA first. If a LoRA is required, apply it against the BF16 transformer and re-quantize, or use the fp8_scaled file on RTX 40xx+ where LoRA composition is best tested.
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-1.1_transformer_only_int8_convrot.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-1.1_transformer_only_int8_convrot.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 1.1 INT8 convrot (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