Ideogram 4.5: readable type and one-spot photo edits

On 30 Sep 2026 Ideogram called 4.5 its most precise edit model in their own post: further rounds should not pile on crumbs, pixel shift, or color drift. In the ChatImg picker the name is Ideogram 4.5.

It does two jobs. A prompt with no photo makes a new image, which is the path for posters and logos whose letters have to match what you typed. A photo you upload is an edit: the part you name changes, and the pixels you did not name are put back from the original.

The product page is Ideogram 4.5. This link opens the picker already on that model: ChatImg.

An existing model-comparison graphic, so a new image and an edit of a photo you already have sit on the same table. Not a screenshot of Ideogram 4.5.

When to pick it

What you have	What you want	Use Ideogram 4.5?
A line of copy and no photo	A poster, wordmark, or menu whose letters are the words you typed	Yes. Do not upload a photo
A photo that already works	A new headline, background, or object, and nothing else moved	Yes. Upload that photo
A comparison of ChatImg and Ideogram as products	How each one starts, and whether you can edit your own photo	No. Use Ideogram alternative
A new account with only trial credits	One test image	No. Trial credits do not cover this model

Practical rule: One change per round. Headline, background, and an added object are three sends. Three requests in one prompt get the unnamed pixels redrawn.

An existing prompt-workflow sketch: write the words that must appear, then judge the layout.

Posters: put the exact words in the prompt

The letters are not a style note. They are the string a person should read. Put that string in quotes and say where it sits.

Copy this as a starting point:

Vertical sale poster, warm off-white ground, one large centered line that reads "Member Day", and a smaller line under it that reads "20% off". No extra language, no watermark.

Check the picker still says Ideogram 4.5 before you generate. The general generator is AI Image Generator. That page does not pin this model.

Practical rule: Read the quoted words before you judge the layout. Wrong letters mean the poster is wrong, even if the grid is clean.

An existing edit-workflow sketch: one change per round, and the rest stays with the upload.

Editing a photo you already have

After the upload, the prompt should describe the change, not restate the whole picture. A full restatement invites a redraw.

Copy this as a starting point:

Change only the headline at the top to "Member Day". Keep the face, the product, the background, and every other line of text as they are.

The check is one question: do the pixels you did not mention still match the upload? Frayed edges, a drifted skin tone, or a rewritten caption next to the headline means this round failed.

Ideogram’s post says sending the result back for another pass should not stack those problems. The next prompt still adds one change, such as the background, and does not repeat the headline you already set.

Practical rule: If an unnamed pixel moved, a leaderboard score does not pass the image.

An existing model-comparison cover. The precise-edit wording follows Ideogram’s 30 Sep 2026 post.

35 credits an image

One image is 35 credits. Square, portrait, and landscape are the same price. A larger canvas is not a second charge.

A new account’s trial balance is 20 credits, and it only spends on models priced at 15 credits or less. Ideogram 4.5 is outside that range. The account needs credits already on it.

This is not the alternative page

People searching for an Ideogram alternative are comparing two products. That page is still Ideogram alternative.

This piece only answers how Ideogram 4.5 in the picker makes a new image with readable type, and how it changes one named spot on a photo you already have. The wording of the precise-edit claim comes from the 30 Sep 2026 post. We do not assign it a leaderboard rank.