Replicate ideogram-ai/ideogram-v2-turbo

If you need legible text inside a generated image - a logo, a poster, a mockup with actual readable words - Ideogram is the name that comes up, and has for years. This node calls the v2 Turbo endpoint, the fast tier of Ideogram v2, through Replicate. Turbo trades a little quality for speed, which suits the way people actually use these: iterate on a headline or a layout until the words land right, then stop.

This is a closed, hosted model. The weights don't exist on your disk and can't - Ideogram kept v1, v2, and v3 fully proprietary. Running it means a Replicate account, a REPLICATE_API_TOKEN, and paying per generation. That's the deal for reaching a model you can't run locally at all.

How it works

Ideogram has arguably the deepest research pedigree in this space - its founders wrote foundational diffusion papers at Google Brain - and text rendering has been its differentiator since day one. Through v3, if you needed accurate text in an image, Ideogram was the benchmark and it was closed. The node is a thin bridge: your prompt and options go to Ideogram's API via Replicate, an image comes back. Note this is the older v2 line, not the local 4.0 weights that opened in 2026 - same lab, different, cloud-only model.

The inputs and outputs that matter

• prompt (required) - describe the scene, and put any text you want rendered in quotes. That's the whole reason you're here.
• aspect_ratio (1:1 default; 16:9, 3:4, and more) - the simple way to set shape. There's also a much more granular resolution enum (dozens of exact pixel sizes) if you need a specific dimension.
• magic_prompt_option (Auto / On / Off) - Ideogram's automatic prompt-expansion. Leave it on Auto to let it embellish; set Off when you want it to render exactly what you typed and nothing more.
• style_type (None / Auto / General / Realistic / Design / Render 3D / Anime) - a quick aesthetic steer.
• negative_prompt, image + mask (for editing an existing image), and seed round it out. force_rerun forces a fresh result.

Output is a single IMAGE.

How to install it

One install covers every Replicate node.

• ComfyUI Manager: search ComfyUI-Replicate, install, restart.
• Manual:

  cd ComfyUI/custom_nodes
  git clone https://github.com/replicate/comfyui-replicate
  cd comfyui-replicate
  pip install -r requirements.txt
  

  then restart.

Nothing lands on your GPU - Ideogram runs on Replicate's servers, so the only install is the replicate Python client. Set your token before launching ComfyUI: export REPLICATE_API_TOKEN="r8_..." (replicate.com/account/api-tokens). No token, no runs.

Common issues & troubleshooting

Auth error on the first run. REPLICATE_API_TOKEN wasn't set before ComfyUI launched. Set it, restart.

The text came out garbled anyway. Keep the text short, put it in quotes, and don't ask for a paragraph - even the text king struggles with long strings. Turning magic_prompt_option to Off helps when the auto-expansion is rewriting your wording out from under you.

It ignored part of my prompt. With magic_prompt_option on Auto, Ideogram may reinterpret. Set it Off for literal adherence.

A prompt got refused / came back odd. Ideogram runs content filtering server-side; some prompts just won't generate. That's a property of the hosted service, not the node.

Costs add up. Every generation is a billed Replicate prediction. Turbo is the cheaper, faster tier, but it's still metered - iterate deliberately.