comfyui-mcp/packs/wan-longer-videos-t2v-96gb/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: wan-longer-videos-t2v-96gb
display_name: WAN 2.2 Longer Videos — Text to Video (96GB fp16)
family: wan
kind: video
description: >-
  WAN 2.2 longer-video TEXT-TO-VIDEO pipeline — full fp16 A14B (High/Low Noise)
  with a video-extend chain that stitches successive clips into longer sequences.
  Uses the FULL fp16 UNets via UNETLoader (not GGUF) + umt5-xxl text encoder,
  Lightning 4-step LoRAs for fast sampling and a FusionX LoRA. The extend chain
  stitches with the I2V models, so both T2V and I2V A14B UNets/LoRAs are
  included — FOUR fp16 UNets at 26.6GB each. Single-purpose graph sliced from
  the wan-longer-videos monolith (one pipeline, no group toggles). This is the
  96GB tier, not a 24GB one. Use wan-longer-videos-t2v for the GGUF quants.
vram: "96GB (full fp16 A14B, 26.6GB per UNet, --highvram); OOMs on 24GB — use wan-longer-videos-t2v for GGUF"
workflow: workflow.json
skill: wan-t2v-video
# WAN 2.2 A14B GGUF video generation is VRAM-hungry; reserving VRAM avoids OOM
# when the GGUF UNet + VAE decode run back to back.
launch_args:
  - "--reserve-vram"
  - "2"
sources:
  mirror: https://huggingface.co/Aitrepreneur/FLX
  installer: WAN2_2-ULTRA-MODELS-NODES_INSTALL.bat
  wan: https://huggingface.co/Wan-AI
notes:
  - Run the generated installer from your ComfyUI root (the folder containing
    custom_nodes/ and models/).
  - Sliced + un-bypassed from the monolith's "WAN 2.2 LONGER TEXT TO VIDEO".
  - This variant installs the FULL fp16 diffusion models
    (diffusion_models/wan2.2_{t2v,i2v}_*_14B_fp16.safetensors, 26.6GB each) and
    loads them with UNETLoader. Launch ComfyUI with --highvram. For anything
    smaller than a 96GB card, install wan-longer-videos-t2v instead — it ships
    the GGUF quants (Q4_K_S <12GB, Q5_K_S 12-24GB, Q8_0 24GB+) and
    UnetLoaderGGUF nodes.
  - "Torch not compiled with CUDA fix: if generation errors with \"Torch not
    compiled with CUDA enabled\", a CPU-only torch was installed. Reinstall the
    CUDA build into the ComfyUI python (portable build):
    python_embeded\\python.exe -m pip install --force-reinstall torch torchvision
    torchaudio --index-url https://download.pytorch.org/whl/cu124"
post_install:
  - Restart ComfyUI (or use ComfyUI-Manager -> Install missing custom nodes for
    any node Python deps).
  - Load workflow.json.