comfyui-mcp/packs/ltx-2.3/pack.yaml

Local-first, agent-native control plane for ComfyUI — MCP server + autonomous sidebar agent that drives your live graph in natural language on ANY LLM: Claude/ChatGPT/Gemini on your subscription (no API key), free local models via Ollama (fully offline),

name: ltx-2.3
display_name: LTX-2.3 Video
family: ltx
kind: video
description: >-
  LTX-2.3 — Lightricks' 22B DiT video model, distributed as a GGUF UNet quant
  for text-to-video and image-to-video. Generates synchronized video + audio
  (separate video and audio VAEs), with a Gemma-3-12B text encoder and a
  text-projection module, an x2 spatial latent upscaler, and distilled speed /
  IC-detailer LoRAs. Drive it from the LTX Director timeline editor in ComfyUI.
vram: "8GB+ (Q4_K_S <12GB, Q5_K_S 12-16GB, Q8_0 24GB+)"
workflow: null
skill: ltxv2-video
launch_args: ["--reserve-vram", "10", "--cache-none"]
sources:
  patreon: https://www.patreon.com/Aitrepreneur
  model: https://huggingface.co/Lightricks
  nodes: https://github.com/Lightricks/ComfyUI-LTXVideo
  mirror: https://huggingface.co/Aitrepreneur/FLX
notes:
  - Run the generated installer from your ComfyUI root (the folder containing
    custom_nodes/ and models/).
  - The UNet is a GGUF quant of ltx-2.3-22b-dev — keep only the quant matching
    your VRAM (Q4_K_S <12GB, Q5_K_S 12-16GB, Q8_0 24GB+); requires ComfyUI-GGUF.
  - OOM mitigation — the documented launch args are
    "--reserve-vram 10 --cache-none". On low-VRAM cards also try a smaller quant.
  - The RunPod installer pins ComfyUI-LTXVideo to commit
    cd5d371518afb07d6b3641be8012f644f25269fc and a torch 2.4.0 / cu121 stack
    (torchvision 0.19.0, torchaudio 2.4.0, xformers 0.0.27.post2,
    transformers 4.51.3). The Windows installers track the node's main branch;
    pin only if you hit workflow incompatibilities.
  - REQUIRED FIX after install — ComfyUI-LTXVideo breaks on kornia 0.8.3+.
    Run fix-ltxvideo-kornia.bat (bundled in this pack) from your
    ComfyUI_windows_portable folder, or apply the patch in post_install below.
post_install:
  - Restart ComfyUI (or use ComfyUI-Manager -> Install missing custom nodes for
    any node Python deps).
  - >-
    Apply the kornia fix to
    custom_nodes/ComfyUI-LTXVideo/pyramid_blending.py (bundled as
    fix-ltxvideo-kornia.bat). kornia 0.8.3+ no longer exports `pad` from
    kornia.geometry.transform.pyramid, so the node fails to import. The patch
    (1) removes the now-broken `pad,` line from the
    `from kornia.geometry.transform.pyramid import ( ... )` block, and
    (2) inserts `pad = F.pad` right after `import torch.nn.functional as F`
    (a .bak_kornia_fix backup is written first). Re-run is idempotent.
  - Load your LTX-2.3 workflow (no workflow.json ships with this pack — import
    one from the LTX Director / ComfyUI-LTXVideo examples).