Ideogram V4

Ideogram V4 is the current text-to-image node for the model that, for a very long time, had exactly one unbeatable superpower: rendering text correctly inside generated images. Posters, logos, signage, UI mockups - stuff that makes every other model produce mush. This node wraps Ideogram 4.0 as a cloud API call, so you get that capability inside a ComfyUI graph without ever downloading weights. It joined core in June 2026, right as Ideogram 4.0 opened its weights - and the KB's Ideogram panel has the full soap opera of what happened after (hint: it involves a safety filter, a licensing ladder, and the community moving on to Krea 2 anyway).

No install, no model files. Same partner-node deal as the rest of this family: link a Comfy account, buy credits, and every generation is a paid API call through Comfy's proxy.

How it works

It's a text-to-image node, full stop - no mask editing here, that's V3's job. You give it a prompt, it posts to Ideogram's ideogram-v4/generate endpoint, and it returns a regular IMAGE tensor you can save or wire into anything downstream. The node keeps the interface deliberately small, which is a relief after V3's sprawling options.

The inputs that matter

• prompt - the text you want rendered as an image. If you're here for Ideogram's famous trick, this is where you put the words you want to see in the image, not just describe.
• resolution - the biggest input, and the one to actually care about. Defaults to Auto, but the dropdown is a curated list of exact pixel sizes like 2048x2048 (1:1) or 2560x1440 (16:9), so you can generate at a precise canvas instead of guessing. Set it once, and the node handles the rest.
• rendering_speed - DEFAULT, TURBO, or QUALITY. Same tradeoff as V3, same pricing ladder: TURBO is the cheap fast lane, QUALITY is the slow expensive one. The price badge on the node tracks it.
• seed - and here's the partner-node disclaimer that appears on every node in this family: it only decides whether the node re-runs. The result is nondeterministic regardless of what seed you set.

Output is IMAGE - a batch if you ask for it, but this node is single-image by design.

Where people get burned

• Expecting V3's editing here. It's not in this node. If you want inpainting or character reference, that's the Ideogram V3 node; V4 is generation only.
• The "Auto" resolution habit. Leaving it on Auto is fine, but the explicit size list is the feature - you lose nothing by picking an exact canvas, and you gain predictable dimensions for layout work.
• The safety-filter legacy. Ideogram's filter is trained into the weights (that infamous refusal image the ComfyUI lead developer had to publicly disclaim). Via the API it's conditional, not absolute, but if you're pushing boundaries expect pushback. The licensing chills are real too: the open-weights terms are non-commercial, and NSFW tuners have reportedly stayed away for fear of takedown. The KB documents all of it.
• Cost. QUALITY speed roughly triples the per-image price versus TURBO. For a single polished piece that's fine; for iterating on twenty variants, stay on TURBO until you've got the prompt right.