Ideogram

Ideogram makes the Ideogram image models, known for readable lettering and layout. Brut calls their own API with your key, for Ideogram 4.5 in two forms:

• Ideogram 4.5, in every image menu (Text → Image, Image → Image, Inpaint, Compose, Multi-Shot, the Storyboard's keyframes, the Sketchpad's generation layers, the workflow cards' Model knobs): text-to-image at 1K or 2K, and an edit that re-renders the picture with your instruction.
• Ideogram 4.5 Precise Edit, on the edit menus (Image → Image, Inpaint, Compose, Annotate & Correct): the picture comes back at its exact size, and every pixel the edit does not touch is copied from it. It is the model's own claim, measured below, and the reason to try it on corrections: several passes on one picture without the drift every re-render adds.
• Ideogram 4.0 Transparent, in every text-to-image menu: a picture on a transparent background, a PNG with alpha, for cutouts, entourage and lettering to place on a layer.

#Get a key

Log in on ideogram.ai, open the API dashboard, create a key under API Keys (the full key is shown once, when it is created) and paste it into ⚙ Settings → Ideogram.

Billing is prepaid credits on that account: one-time amounts from $10, or an auto-recharge; a key pauses when the balance reaches $0, and if you hold an Ideogram subscription its monthly credits are spent first. There is no free tier. No aggregator carries Ideogram, so the key is the only way in.

#Models

The prices are Ideogram's own: a quality costs the same at 1K and at 2K, at the picture's own size, with a mask and with reference pictures. Masks and references add nothing. The node shows the price before a run and the ledger row carries it after.

Things worth knowing:

• Your prompt is sent as written. Ideogram can rewrite and expand prompts on its side ("magic prompt"); Brut turns that off, so what the node says is what the model gets.
• Sizes. On text-to-image, a 16:9 picture comes back at 1280 × 720 at 1K and 2560 × 1440 at 2K, a square at 1024 or 2048. 21:9 exists at 2K only (3072 × 1280): at 1K the ratio is greyed in the menu with the reason, and a node already set to it shows a line with the two ways out. On an edit, Same as the picture returns the picture at its own size (a picture too large for the model comes back scaled down, proportions kept); a picked size with aspect auto keeps the picture's proportions at that size, and a picked ratio reshapes it.
• Masks are real masks here. On the Inpaint node, Ideogram paints inside the mask and keeps the rest; the result is always the picture's own size, so the Size menu must stay on Same as the picture beside a mask (the run says so otherwise). On Precise Edit, the pixels outside the edit are copied from the original, mask or not. Measured on a 2048 × 1143 interior with one instruction, the sofa recolored to green velvet: the result came back at 2048 × 1143 with 95 percent of its pixels identical to the input, byte for byte, and every changed pixel inside the sofa's part of the frame. That is the behavior every other edit model only approximates, and the reason to try it first on a correction.
• References are references. On Compose and Image → Image, the pictures wired after the first guide the edit and are never edited themselves.
• The quality menu is the model's. Text-to-image offers low, medium and high; both edits add very_low, the fastest and cheapest, for a first look. In our runs a text-to-image took about 7 seconds, a precise edit about 9.
• On Annotate & Correct, one pass per mark. Pick Ideogram and the Precise Edit on the Annotate node and Brut runs one precise edit per mark, in marker order, each on the previous pass's result: a box, circle or lasso confines its edit with a mask of its shape; an arrow or a point names its object and where it sits in the picture. The picture outside each edit is copied as it is, so five marks leave a picture that differs from the original only where you marked. Five marks cost five passes at the quality you picked (5 × $0.03 at low). The drawn-marks approach every other model uses came back here with the markers painted into the result, which is why the marks travel this way on this model.
• Pictures by Ideogram. Every picture these models make is generated by Ideogram's models, as their terms ask every app to say.

#Ideogram 4.0 Transparent

A text-to-image that comes back as a PNG with alpha: the subject alone, nothing behind it. It is the one for cutouts and entourage (a tree, a chair, a person, a car), logos and lettering to place on a layer. In every text-to-image menu.

A cutout keeps its transparency wherever Brut composites it: on a Layers node it shows the stack below through its transparent pixels, with or without a mask, and in the Sketchpad's tray it places as an image layer with nothing around the subject.

#Limits

• Pictures in: JPEG, PNG or WEBP up to 25 MB each (other common formats work too, with no quality lost), with an aspect ratio between 1:6 and 6:1.
• Results are saved to your project as soon as they're ready.
• Ideogram's safety check can refuse a prompt (nothing is charged) or withhold a result; the node shows which.
• Ideogram runs ten requests at a time per account by default; past that, runs wait their turn instead of failing.

#Privacy

Calls go from your machine to Ideogram's servers with your key, and nowhere else. Ideogram's own terms govern what they keep of the pictures you send and receive.