A clean and minimal reference workflow for Ideogram 4 in ComfyUI.
This workflow demonstrates the core Ideogram 4 inference pipeline without additional helper or UI nodes. It is intended as a starting point for learning, experimentation, and building your own workflows.
Tested with:
ComfyUI 0.27.0
Ideogram 4 FP8 (Conditional & Unconditional UNETs)
Flux2 VAE
Qwen3VL-8B-Uncensored-HauhauCS GGUF (
Type = ideogram4)NVIDIA RTX 2080 (8 GB VRAM)
Note: This workflow uses CLIPLoader (GGUF) with Type = ideogram4. Do not replace it with DualCLIPLoader or change the type to flux, as the workflow will fail.
Description
Version 1.0
Created from scratch after studying the official Ideogram4 workflow.
Purpose
✓ Educational
✓ Minimal
✓ Easy to extend
Verified
✓ Text-to-Image
Next
□ Prompt Builder
□ LoRA
□ Image-to-Image
□ ControlNet
Comments (2)
Nice workflow - separated loaders are really useful. but didn't get what "Anything Everywhere" node is needed for, deleted it (not to install custom node pack).
So, i've added "(Deno) Ideogram Director" + "Preview as Text" nodes from Deno's workflow and two "Lora Loader Stack (rgthree)" nodes for cond and uncond models + tryied "CLIPMergeSimple" node for their CLIP outputs - but it didn't work, so i used the first Lora CLIP output only.
Worked well with ig4 fp8 scaled amd int8 convrot as well, 3 Loras and with Q6 CLIP model.
16gb vram, 32gb ram
Cool.. that is the idea :) Feel free to modify the workflow according to your needs.. Much appreciate the kind words.