comfyui-mcp/packs/z-image-xy-plot/manifest.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),

# Z-Image Turbo XY-Plot (LoRA comparison) install manifest — consumable by apply_manifest (src/services/manifest.ts).
# Ground truth: XY-PLOT-LORA-Z-IMAGE-TURBO-MODELS-NODES_INSTALL.bat
#   (cross-checked: XY-PLOT-LORA-Z-IMAGE-TURBO-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat, XY-PLOT-LORA-Z-IMAGE-TURBO-AUTO_INSTALL-RUNPOD.sh).
# Reconciled against XY-PLOT-LORA-Z-IMAGE-WORKFLOW.json (workflow = source of truth).
# Mirror base: https://huggingface.co/Aitrepreneur/FLX/resolve/main
# Official Z-Image Turbo weights: https://huggingface.co/Tongyi-MAI/Z-Image-Turbo
#
# GGUF QUANT TIER: the installer prompts Q5_K_S (GPUs <8GB) / Q6_K (8-12GB) /
# Q8_0 (12-16GB+, best quality) for the Z-Image Turbo UNet. This manifest ships
# the Q8_0 tier active, which is what the bundled workflow's UnetLoaderGGUF
# selects and what every other z-image-* pack ships. Until #1767 this pack alone
# activated Q5_K_S while the workflow asked for Q8_0, so a fresh install opened
# to a missing-model error. To run a smaller tier, comment the Q8_0 line below,
# uncomment the tier you want, and pick it in the UnetLoaderGGUF node.
#
# This pack compares USER-TRAINED Z-Image LoRAs on an XY grid. The LoRAs the
# workflow references (the YOUR-LORA-*.safetensors placeholders, declared in
# pack.yaml external_models) are user-supplied and are NOT downloaded here —
# drop your own trained LoRAs into models/loras/ and select them in the
# "easy loraNames" nodes.

pip: []

custom_nodes:
  - https://github.com/city96/ComfyUI-GGUF
  - https://github.com/rgthree/rgthree-comfy
  - https://github.com/yolain/ComfyUI-Easy-Use
  - https://github.com/kijai/ComfyUI-KJNodes
  - https://github.com/TinyTerra/ComfyUI_tinyterraNodes
  # --- Shared Z-Image Turbo node bundle (cloned by the installer; not used by
  #     this XY-plot workflow, but kept for parity with the base Turbo pack) ---
  - https://github.com/ssitu/ComfyUI_UltimateSDUpscale
  - https://github.com/cubiq/ComfyUI_essentials
  - https://github.com/wallish77/wlsh_nodes
  - https://github.com/ClownsharkBatwing/RES4LYF

models:
  # --- Text encoder (CLIPLoaderGGUF, lumina2) ---
  - url: https://huggingface.co/Aitrepreneur/FLX/resolve/main/Qwen3-4B-UD-Q6_K_XL.gguf
    local_path: text_encoders/Qwen3-4B-UD-Q6_K_XL.gguf

  # --- VAE (VAELoader) ---
  - url: https://huggingface.co/Aitrepreneur/FLX/resolve/main/ae.safetensors
    local_path: vae/z-image-ae.safetensors

  # --- Diffusion UNet (UnetLoaderGGUF) — Q8_0 tier active (matches the workflow) ---
  - url: https://huggingface.co/Aitrepreneur/FLX/resolve/main/z_image_turbo-Q8_0.gguf
    local_path: unet/z_image_turbo-Q8_0.gguf
  # Q5_K_S tier (<8GB) — uncomment to use, comment the Q8_0 line above
  # - url: https://huggingface.co/Aitrepreneur/FLX/resolve/main/z_image_turbo-Q5_K_S.gguf
  #   local_path: unet/z_image_turbo-Q5_K_S.gguf
  # Q6_K tier (8-12GB) — uncomment to use, comment the Q8_0 line above
  # - url: https://huggingface.co/Aitrepreneur/FLX/resolve/main/z_image_turbo-Q6_K.gguf
  #   local_path: unet/z_image_turbo-Q6_K.gguf

  # --- Upscalers (in the installer's shared Turbo bundle; NOT used by this
  #     XY-plot comparison workflow — uncomment if you want them) ---
  # - url: https://huggingface.co/Aitrepreneur/FLX/resolve/main/4x-ClearRealityV1.pth
  #   local_path: upscale_models/4x-ClearRealityV1.pth
  # - url: https://huggingface.co/Aitrepreneur/FLX/resolve/main/RealESRGAN_x4plus_anime_6B.pth
  #   local_path: upscale_models/RealESRGAN_x4plus_anime_6B.pth

  # --- LoRAs: NONE bundled. Bring your own trained Z-Image LoRAs into
  #     models/loras/ — this workflow compares them on an XY grid. ---