Ideogram V3

Ideogram V3 is the Swiss-army node of the Ideogram family in ComfyUI: it generates images from text, it edits existing images with a mask, and it does character-reference generation - all through one node and one cloud API call per run. If the whole point of Ideogram is accurate text rendered inside images, this is the node that gives you the full toolkit for it, no local model, no VRAM, just a prompt and a Comfy account with credits.

It ships with ComfyUI core as a partner node, so there's nothing to install. It's the model the community actually knows: Ideogram stayed closed through three generations while its founders' diffusion pedigree went mostly unnoticed, and its text-in-image quality was the benchmark nothing open matched - the KB's Ideogram panel has the whole story. V3 is that reputation, wrapped in a node.

How it works

Three modes, one node:

• Text-to-image: prompt alone → image. You can optionally attach a character_image (plus a character_mask) so the output keeps a specific person consistent across generations - the character-reference mode.
• Inpainting: connect both image and mask. White areas of the mask get replaced, per the tooltip. If you only connect one of the two, it errors on purpose - the node needs both to know what to edit.
• The node posts to Ideogram's generate or edit endpoints through Comfy's proxy and returns a normal IMAGE tensor.

The inputs that matter

• prompt - the text you're turning into an image, or the description of the edit.
• image + mask - the editing pair. White mask = replaced.
• character_image + character_mask - consistency for people. Feed it a reference, keep the same face across shots.
• resolution vs aspect_ratio - this is the subtle one. resolution defaults to Auto; if you set it to a concrete value it overrides aspect_ratio, which is otherwise ignored. Set one, not both, unless you know exactly what you want.
• rendering_speed - DEFAULT, TURBO, or QUALITY. The classic tradeoff, and it's priced differently: TURBO is cheaper, QUALITY costs more. The price badge reflects it.
• magic_prompt_option - AUTO/ON/OFF. Ideogram's prompt-rewriting helper; AUTO lets it decide.
• num_images - up to 8. Each one is a separate charge, so watch the badge.
• seed - the usual partner-node caveat: it only controls whether the node re-runs; the result is nondeterministic regardless.

Output is IMAGE, batched if you asked for more than one, ready for Save Image or whatever comes next.

Where people get burned

• Half-connected edit inputs. Image without mask (or mask without image) is a deliberate error. The node tells you what it needs.
• Mask polarity. White = replaced here. If your mask came from a workflow that uses the opposite convention, flip it with Invert Mask before feeding it.
• Resolution overriding ratio. Set a fixed resolution and your carefully chosen 16:9 ratio silently stops mattering.
• Cost. Character reference and QUALITY mode push the per-image price up, and num_images multiplies it. The badge shows it all - glance at it before you queue a batch of eight.