comfyui-mcp/packs/ernie/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: ernie
display_name: ERNIE-Image Ultra
family: ernie
kind: txt2img
description: >-
  ERNIE-Image Ultra — Baidu's open-weight (~8B single-stream DiT, Apache-2.0)
  TEXT-TO-IMAGE model, repackaged for ComfyUI by Comfy-Org. Standout strength is
  precise multilingual TEXT RENDERING (including Chinese) — posters/signage/UI
  mockups, manga/anime multi-panel layouts — plus strong prompt/instruction
  following for complex multi-object scenes. This pack uses the distilled Turbo
  GGUF, which is high-fidelity in ~8 steps at cfg 1 and runs in <8GB VRAM.
  Not an instruction-based image editor (use qwen-image-edit / Flux Kontext for
  "change X in this photo").
vram: "<8GB"
workflow: workflow.json
skill: ernie-image
launch_args: []
# User-supplied example LoRAs referenced by the (bypassed) Power Lora Loader.
# Declared external so packs:check-models knows they're intentionally not shipped.
external_models:
  - hirohiko-araki-style-ERNIE_000001250.safetensors
  - ernie-anime-v1.safetensors
sources:
  model: https://huggingface.co/Comfy-Org/ERNIE-Image
  original: https://github.com/baidu/ERNIE-Image
  docs: https://docs.comfy.org/tutorials/image/ernie-image/ernie-image
  mirror: https://huggingface.co/Aitrepreneur/FLX
notes:
  - Run the generated installer from your ComfyUI root (the folder containing
    custom_nodes/ and models/).
  - Text-to-image only — ERNIE excels at literal text/typography (write the exact
    string you want in quotes) and structured layouts. It is NOT an instruction
    editor.
  - Turbo defaults (from the shipped graph) — 8-9 steps, cfg 1, euler/simple,
    denoise 1.0, with ModelSamplingAuraFlow shift=3.1 applied before sampling.
  - Quant guidance — Q5_K_S <8GB · Q6_K 8-12GB · Q8_0 12-16GB+. The manifest
    ships Q8_0; edit the filename to pull a smaller quant.
  - COMBO PIPELINES — 10 of the 20 groups chain ERNIE with Z-Image-Turbo
    ("ZIT ---> ERNIE", "TWO TIMES COMBO ERNIE ZIT", etc.), so the manifest also
    fetches the Z-Image-Turbo models (z_image_turbo-Q8_0.gguf, Qwen3-4B, and the
    z-image VAE saved as z-image-ae.safetensors). Pure-ERNIE groups use
    ernie-image-turbo-*.gguf + ministral-3-3b + flux2-vae (+ optional enhancer);
    the combo groups ship bypassed — toggle one on via the rgthree group
    bypassers, and expect higher VRAM since both models load.
  - Weights are mirrored from the third-party huggingface.co/Aitrepreneur/FLX
    repo; prefer the official Comfy-Org/ERNIE-Image repo (same filenames) when
    possible.
post_install:
  - Restart ComfyUI (or use ComfyUI-Manager → Install missing custom nodes for
    any node Python deps).
  - Load workflow.json.
  - If TextGenerate (prompt enhancer) shows as missing, install it via
    ComfyUI-Manager, or set the ComfySwitchNode to the raw prompt (switch=false).