name: ideogram-txt2img
display_name: Ideogram 4 — Text to Image
family: ideogram
kind: txt2img
description: >-
  Ideogram 4 (Ideogram Ultra) text-to-image, running LOCALLY from the Comfy-Org
  fp8 DiT weights (NOT the Ideogram cloud API). Pairs a Qwen3-VL text encoder
  with a Gemma4 prompt-builder LLM (JSON Prompt Builder subgraph) and a dual
  conditional/unconditional UNet guider for strong prompt adherence and crisp
  TEXT RENDERING. Single-purpose txt2img graph sliced from the ideogram monolith
  (one pipeline, no group toggles), with optional KJ image sharpening and an
  off-by-default Power Lora Loader.
vram: "<8GB"
workflow: workflow.json
skill: ideogram-txt2img
launch_args: []
sources:
  model: https://huggingface.co/Comfy-Org/Ideogram-4
  mirror: https://huggingface.co/Aitrepreneur/FLX
notes:
  - Run the generated installer from your ComfyUI root (folder with custom_nodes/
    and models/).
  - Local weights only — this does NOT call the Ideogram cloud API; the fp8
    Ideogram-4 DiT is loaded via UNETLoader and sampled with SamplerCustomAdvanced.
  - Sliced + un-bypassed from the monolith's "TEXT TO IMAGE" group.
  - Ideogram excels at literal text/typography — write the exact string in quotes.
post_install:
  - Restart ComfyUI (or ComfyUI-Manager -> Install missing custom nodes for any
    node Python deps).
  - Load workflow.json.