BYOKey Ideogram Image

Through its V3 generation, Ideogram was the benchmark for text rendering in generated images - logos, signage, marketing copy, anything with letters that need to be spelled correctly. The KB's history is clear: nothing open matched its text-in-image accuracy for years. BYOKey_Ideogram calls the Ideogram V3 API (api.ideogram.ai, key sent as the Api-Key header) with your own key and returns a ComfyUI IMAGE - no Comfy proxy, no credits, your key against Ideogram directly.

(Yes, Ideogram 4 later went open-weight, which shifted the community's center of gravity - but the hosted V3 API is still the version this node talks to, and the open weights are non-commercial anyway. If you want clean text in an image without licensing questions, the API is the no-surprises route.)

How it works

It POSTs your prompt to Ideogram's V3 endpoint and decodes the returned image into an IMAGE tensor. The interesting part is the controls: Ideogram is one of the few image APIs where you get explicit levers for the text-rendering and design aspects that are its whole identity.

The inputs that matter

• prompt + negative_prompt - the generation request, with a real negative prompt field (a rarity on hosted image APIs).
• aspect_ratio vs. resolution - this pairing confuses people, so read carefully: resolution defaults to Auto, and only when it's Auto does aspect_ratio do anything. The moment you pick a concrete resolution (and there are ~70 choices, from 512x1536 up), it overrides aspect_ratio entirely. Rule of thumb: leave resolution on Auto and just pick your ratio.
• style_type - AUTO/GENERAL/REALISTIC/DESIGN/FICTION. The DESIGN mode is where Ideogram's text-and-layout strengths live; FICTION pushes a more illustrated look.
• magic_prompt_option - AUTO/ON/OFF controls Ideogram's automatic prompt enhancement. ON makes the model expand your prompt for better results but less predictability.
• rendering_speed - DEFAULT/TURBO/QUALITY, the speed/quality dial.
• num_images (1–8) and seed (0 = unset) - batch count and reproducibility.

Output is output (IMAGE) - standard wire-in anywhere.

Installing it

It's one of the 37 nodes in comfyui-byokey, with no extra dependencies beyond what ComfyUI bundles (torch, aiohttp, PIL, numpy, av). Install via ComfyUI Manager (search "byokey") or:

cd ComfyUI/custom_nodes
git clone https://github.com/MeteorAndy/comfyui-byokey.git

Restart, and it's under api/byokey/image. No model downloads - Ideogram does the compute.

Gotchas

Two practical traps. First, the aspect_ratio/resolution override behavior above - people set a ratio, wonder why it's ignored, and it's because a resolution is still selected. Set resolution to Auto and your ratio starts working. Second, MagicPrompt and QUALITY speed both burn more credits per call without necessarily being what you wanted; leave MagicPrompt on AUTO and only bump rendering_speed when you're actually unhappy with results. The standard pack note: your key is a widget value saved into the workflow JSON - scrub before sharing, and remember this paste-your-key category has a documented malware history, so reading the open source is the sane habit. If your workflow is logos, posters, or anything with legible text, this is the hosted image node to reach for.