SageAttention is not new enough version — LTX-2
SageAttention is not new enough version or could not determine CUDA architecture, cannot apply LTX2 memory efficient sage attention patch. The workflow can still run; this page covers LTX-2, LTX-2.3, and LTX-2.5 setup.
LTX-2 is Lightricks' open-weights audio-video model. ComfyUI supports the family natively through the template library. You do not need a separate wrapper custom node for the official templates.
If you searched sageattention is not new enough version or could not determine cuda architecture, cannot apply ltx2 memory efficient sage attention patch, that is a performance warning, not a missing-module failure. The workflow can still run. For No module named 'sageattention', use the SageAttention missing-module page.
SageAttention is not new enough version
That log means the LTX-2 memory-efficient SageAttention patch did not apply. ComfyUI usually continues with standard attention. Update SageAttention only if you want the patch; switch the node to sdpa or comfy if you just want a clean queue. Setup, model folders, and the rest of the LTX-2 family live on this page.
LTX-2 is a 19B DiT audio-video model: it generates motion, dialogue, sound effects, and music in one pass. Later official releases keep that joint audio-video path and add quality, encoder, and decoder upgrades. Do not mix files from LTX-2, LTX-2.3, and LTX-2.5 in the same workflow.
Fast answer
Update ComfyUI, open Template Library → Video, load an LTX-2, LTX-2.3, or LTX-2.5 template, and follow the pop-up to download the files it asks for. If you see SageAttention is not new enough version ... cannot apply ltx2 memory efficient sage attention patch, that is a performance warning, not a hard failure — the workflow still runs.
Which LTX version you have
| Release | Official ComfyUI entry | Encoder family | Do not mix with |
|---|---|---|---|
| LTX-2 | Template Library → Video → LTX-2 | Gemma 3 era templates | 2.3 / 2.5 weights |
| LTX-2.3 | Template Library → search LTX-2.3 (ComfyUI 0.16.1+) | Gemma 3, including gemma_3_12b_it_fp4_mixed.safetensors | 2 / 2.5 weights |
| LTX-2.5 | Templates panel → LTX-2.5 (ComfyUI 0.32.0+) | Custom Gemma 4 12B encoder | 2 / 2.3 weights |
The SageAttention patch wording still says ltx2 on later templates. Stay on this page for that log. Partner Node (hosted API) workflows are a different product path and are out of scope here.
What native LTX-2 can do
Official ComfyUI docs list these modes for LTX-2:
| Mode | What you provide | What you get |
|---|---|---|
| Text-to-video (T2V) | A prompt | Video with synchronized audio |
| Image-to-video (I2V) | A start image + prompt | The image is animated |
| Video-to-video | A video + prompt | Restyled or continued motion |
| Canny / Depth / Pose control | Control video plus IC-LoRAs | Structure-guided video |
| Native upscalers | Spatial x2 and temporal x2 checkpoints | Higher resolution or FPS |
LTX-2.5 official notes add native multi-shot generation, a diffusion video decoder, a prompt enhancer, and an optional duration head. Those features still use the same first step: load the matching official template.
Step 1: Update ComfyUI
- LTX-2 templates need a current ComfyUI release with the built-in LTX nodes.
- LTX-2.3 official blog asks for ComfyUI 0.16.1 or newer.
- LTX-2.5 official blog asks for ComfyUI 0.32.0 or Comfy Cloud.
For a Git install:
cd ComfyUI
git pull
python -m pip install -r requirements.txtFor the Windows portable package, run the update from the portable root with the bundled Python. If the LTX nodes are missing after updating, restart ComfyUI and confirm the frontend package updated too (see comfyui-frontend-package Not Installed or Version Outdated).
Step 2: Download the official template
The authoritative file list is the pop-up inside the template, not a blog screenshot.
- Open ComfyUI in the browser.
- Open Workflow → Browse Templates (or the Templates panel).
- Open Video and load LTX-2, or search LTX-2.3 / LTX-2.5.
- Follow the download pop-up. Save each file into the folder the pop-up names.
Official LTX-2 tutorial: LTX-2.0 ComfyUI workflow examples. Official 2.5 weights: Lightricks/LTX-2.5. Official 2.0 weights: Lightricks/LTX-2.
LTX-2 official checkpoints
The official ComfyUI LTX-2 docs and the Lightricks model card name these checkpoints. The template pop-up decides the folder. Do not guess models/ltx_video from third-party posts.
| Checkpoint name | Notes |
|---|---|
ltx-2-19b-dev | Full 19B model, bf16, flexible / trainable |
ltx-2-19b-dev-fp8 | Full model, fp8 |
ltx-2-19b-distilled | Distilled, 8 steps, CFG=1 |
ltx-2-spatial-upscaler-x2-1.0 | 2x spatial upscaler |
ltx-2-temporal-upscaler-x2-1.0 | 2x temporal upscaler |
The Lightricks card also lists ltx-2-19b-dev-fp4 and ltx-2-19b-distilled-lora-384. Use a file only when your loaded template asks for it.
LTX-2.5 official Comfy-aligned folders
Lightricks publishes LTX-2.5 as a split Comfy-aligned pack. Each component is one .safetensors file. Official repo folders map to the same names under your running ComfyUI/models/ tree.
Diffusion models — ComfyUI/models/diffusion_models/
| File | Notes |
|---|---|
ltx-2.5-22b-dev-transformer-bf16.safetensors | Full / trainable DiT |
ltx-2.5-22b-distilled-transformer-bf16.safetensors | Distilled DiT, 8-step schedule, CFG=1 |
ltx-2.5-22b-dev-transformer-comfy-int8-convrot.safetensors | Full DiT, Comfy int8 (ComfyUI only) |
ltx-2.5-22b-distilled-transformer-comfy-int8-convrot.safetensors | Distilled DiT, Comfy int8 (ComfyUI only) |
ltx-2.5-22b-distilled-transformer-nvfp4.safetensors | Distilled DiT, NVFP4 |
Text encoders — ComfyUI/models/text_encoders/
| File | Notes |
|---|---|
gemma4-12b-with-proj-ltx-2.5-bf16.safetensors | Gemma 4 12B encoder plus projections |
gemma4-12b-with-proj-ltx-2.5-comfy-int8-convrot.safetensors | Same encoder, Comfy int8 |
VAE — ComfyUI/models/vae/
| File | Notes |
|---|---|
ltx-2.5-video-vae-bf16.safetensors | Diffusion video decoder, higher quality |
ltx-2.5-video-vae-conv-bf16.safetensors | Conv VAE, faster / lighter |
ltx-2.5-audio-vae-bf16.safetensors | Audio VAE and vocoder |
Other official 2.5 files
| File | Folder | Notes |
|---|---|---|
ltx-2.5-22b-distilled-lora-450-bf16.safetensors | models/loras/ | Distilled LoRA for dev-transformer workflows |
ltx-2.5-duration-head-bf16.safetensors | models/model_patches/ | Optional auto duration |
ltx-2.5-latent-spatial-upscaler-x2-bf16-1.0.safetensors | models/latent_upscale_models/ | x2 spatial upscaler |
ltx-2.5-latent-temporal-upscaler-x2-bf16-1.0.safetensors | models/latent_upscale_models/ | x2 temporal upscaler |
Folder mistakes to avoid
Do not drop a Gemma encoder into checkpoints or a 2.5 transformer into a 2.0 LTX-2 template. LTX-2.3 still uses Gemma 3 files such as gemma_3_12b_it_fp4_mixed.safetensors in models/text_encoders/ — that is not the 2.5 Gemma 4 file. If a dropdown is empty after moving files, restart ComfyUI and confirm you edited the install that is running. See Where to Put Safetensors in ComfyUI.
Step 3: Run the workflow
- Load the matching template and confirm no nodes are red.
- Point each loader at the files you downloaded.
- For I2V or first/last-frame, connect the images the template asks for.
- Write a detailed chronological prompt. Official LTX-2 prompting advice is literal action, camera, lighting, and environment in one flowing paragraph, usually under 200 words.
- Queue. Joint audio-video models output picture and sound together when the audio VAE / vocoder file is present.
Common errors
| Error | What it means | Fix |
|---|---|---|
SageAttention is not new enough version or could not determine cuda architecture, cannot apply ltx2 memory efficient sage attention patch | SageAttention is missing, too old, or CUDA arch was not detected; ComfyUI falls back to standard attention | Usually a warning — the workflow still runs. To silence it, update SageAttention or use sdpa/comfy attention. See No module named 'sageattention' |
No module named 'sageattention' | Optional acceleration backend missing | Not fatal for most queues; switch the node to sdpa, comfy, torch, or auto. See SageAttention missing |
No module named 'triton' | Triton layer missing or wrong for the installed Torch/CUDA | Fix Triton before SageAttention. On Windows use triton-windows, not upstream triton. See Triton missing or unavailable |
Out of memory / Not enough GPU memory | VRAM exhausted | Use the distilled or int8/NVFP4 variant your template lists, lower resolution or frame count, or see CUDA Out of Memory |
| Empty model dropdown | File in the wrong folder, wrong LTX version, or ComfyUI not restarted | Match the table above, restart, and confirm the running install path |
| Missing nodes in a downloaded workflow | ComfyUI is older than the template, or a custom node failed to import | Update to the version the template needs; see Workflow Missing Nodes |
FAQ
Is the LTX-2 SageAttention patch error fatal?
No. sageattention is not new enough version or could not determine cuda architecture, cannot apply ltx2 memory efficient sage attention patch is a fallback warning. ComfyUI continues with standard attention.
Do I need ComfyUI-LTXVideo to run LTX-2?
No for the official native templates. ComfyUI ships built-in LTX workflows. The Lightricks ComfyUI-LTXVideo pack is a separate custom-node path with its own folder notes. If you loaded a Template Library LTX workflow, follow that pop-up, not a random custom-node README.
Where do LTX-2.5 models go in ComfyUI?
Official Lightricks packaging is Comfy-aligned: transformers in models/diffusion_models/, Gemma 4 in models/text_encoders/, VAEs in models/vae/, LoRAs in models/loras/, the duration head in models/model_patches/, and latent upscalers in models/latent_upscale_models/.
Can I reuse LTX-2.3 files in an LTX-2.5 template?
Do not mix the packs. LTX-2.3 uses Gemma 3. LTX-2.5 uses a custom Gemma 4 12B encoder and a new decoder. Lightricks notes that many 2.3 LoRAs still run on 2.5, but the base transformer, encoder, and VAE should stay on the 2.5 set.
Does LTX-2 generate audio?
Yes. Official LTX-2 generates synchronized audio with the video in one pass. For 2.5, the official pack includes ltx-2.5-audio-vae-bf16.safetensors.
Related Guides
- Where to Put Safetensors in ComfyUI — folder map for model files
- SageAttention is not new enough version — MiniMax H3 — the same warning family on H3
- No module named 'sageattention' in ComfyUI — missing-module intent
- Triton missing or unavailable — Triton errors on Windows
- CUDA Out of Memory — VRAM fixes
- ComfyUI Failed to Get Custom Node List — if template downloads fail
Source References
- LTX-2.0 ComfyUI workflow examples (official ComfyUI docs)
- LTX-2 open-source audio-video in ComfyUI (official ComfyUI blog)
- LTX-2.3 day-0 support in ComfyUI (official ComfyUI blog)
- LTX-2.5 day-0 support in ComfyUI (official ComfyUI blog)
- Lightricks/LTX-2 (Hugging Face)
- Lightricks/LTX-2.5 (Hugging Face)
Start free with Wonderful Launcher if this affects your real ComfyUI environment. It keeps launcher-native repair, task logs, and runtime checks in one place; credits are only for image generation and metered tools.
Download Wonderful LauncherSee credit packagesDid this fix your issue?
Your answer helps prioritize verified ComfyUI repairs.