Ideogram 4 in ComfyUI: setup and layout control tutorial

This tutorial covers Ideogram 4 image generation in ComfyUI, using a KJ Prompt Builder workflow to place objects and text with bounding boxes. Examples include posters and manga panels, with controls for typography, colors and individual body parts. The presenter prefers its composition control and prompt adherence to Z-image and Flux Klein, but reports roughly a minute per image on their setup.

The installation walkthrough uses ComfyUI's Windows portable version. It covers ComfyUI Manager, missing custom nodes and manual installation of KJ nodes. The workflow requires both a main and an unconditional model, a text encoder and the Flux 2 VAE. The presenter selects FP8 files and describes NVFP4 alternatives for compatible GPUs. They report that CPU offloading can allow operation with 6 GB of VRAM when enough system RAM is available; this is not a universal hardware guarantee.

The workflow demonstration explains aspect ratios, seeds and batch size. The presenter attributes an apparent safety-filter result to missing bounding boxes in their example. A generated image can serve as a canvas background for arranging boxes, but the demonstrated model accepts text only, not reference images.

A separate sponsored segment describes Higgsfield MCP connectors, an external service distinct from the local workflow. In the closing license discussion, the presenter says Ideogram 4 has a non-commercial license and directs commercial users to sales, despite calling it open source earlier.