Astounding AI Image Update — Ideogram Drops v.4.0!

Perfect Text, Editable Layers, JSON Prompting, Bounding Boxes, Transparency, and More…

banner image for “Astounding AI Image Update — Ideogram Drops v.4.0!” — created by the author using Ideogram v.4.0 and PhotoPea

Ideogram dropped version 4.0 of its image generating model this week and it has gathered a lot of attention. Here are some of its features:

• Industry-Leading Typography — including Editable Text Layers
• Native 2K Resolution — 1:1 aspect ratio is 2048 x 2048 pixels and 16:9 is 2560 x 1440 pixels
• Native Background Transparency

Let’s start with something very special about this upgrade and release!

The Ideogram developers have released this model as “the first open-weight text-to-image” model.

a screen capture from the Ideogram “home” page on Github — https://github.com/ideogram-oss/ideogram4

What does all that technical stuff mean?

You may have heard of software that is “open source” — that means that the developer/programmer has posted the original source code for their software so others may view it, learn from it, and download it for their own computer free of charge. They also can modify the source to add or alter features in the software. (**Github.com** is a website where many developers upload and share their source code.)

And the “open-weight” means that for this model (software) the developers are sharing the data from training the model so that others don’t have to do the training part on their own. This is a big deal for Ideogram to do this! They are sharing their “intellectual property” with the programming community for viewing, using, and collaboration.

If you are interested in this you can read more about it here (**https://ideogram.ai/models/4.0/**).

If not let’s get to the new model, its interface, and its features.

The Ideogram Interface

a screen capture of the Ideogram main page once you are logged in — you can use Ideogram free with free tokens every day — https://ideogram.ai/

This is what the page looks like when you first arrive at https://ideogram.ai/. (Ideogram does have a free tier to try it out with a limited number of “tokens” each day. They also offered a few paid subscription tiers.)

At the center of the page is the welcoming phrase, “What will you create?”, much like other chatbot type image generators. You can begin typing in the box where it says, “Generate new or upload and edit”.

Below the area where you can type your prompt are some buttons to help you. The first one looks like a paperclip and is labeled “add image”. Click to load an image as a reference or to edit.

Next is the number of the model — it shows 4.0 (latest). You can use earlier models if you wish to change it.

Next is the “count and speed” button. It shows the number of images you can create with the prompt, and the speed of the creation. The default number of image is 4, and you can choose “Turbo”, “Default”, or “Quality” as the speed. “Quality” takes more time to create and costs more tokens. Try “Default” or “Turbo” for faster, less expensive generations.

The next button shows the aspect ratios that you can choose.

a screen capture of the aspect ratio drop-down menu

Ideogram does offer a lot of common aspect ratios plus a few modern additions for mobile devices. (There is a custom aspect ratio option but it has some limitations.)

All the way over to the right there is an upward pointing arrow in a black circle. Click that to begin generating your image.

There are more buttons but I think that is enough to get us started.

Ideogram 4.0 Features

The announcement claims that Ideogram creates “best-in-class multi-lingual text rendering”. (For years I have been using Ideogram to create posters and banner like the one below.)

banner for “Ideogram creates flawless text!” — created by the author in Ideogram 4.0

Feature 1: Perfect Text Rendering

Let’s put it to the test and create a simple poster image with a text headline in bold text.

Here is the prompt that I typed into the prompt box, “create the text “Bach Is For Everyone” in a stylish font on a plain white background”. I kept the model set to 4.0, and I set the aspect ratio to 5:4.

Ideogram created 4 “sample” images for me. The one I liked the best is shown below. I captured the right sidebar as part of the image so that you could see some of the action buttons that can be used. (Note: Ideogram added a small graphic element and a subtitle “A Musical Manifesto” which was not part of my prompt”.)

a screen capture of my poster image in the creation page — also showing the right sidebar

The extra items were added by the “Magic Prompt” feature which Ideogram uses to enhance the image prompts.

(Note: In previous versions there was an option where you could disable the magic prompt so that Ideogram wouldn’t add/enhance your prompt unless you wanted it to. I see that in this model the magic prompt is included. I assume they will add a way to turn it off if you don’t want it. At any rate I could edit out the additions if I don’t want to keep them.)

If you are interested here is the enhanced “magic” prompt text.

High level description
A typographic poster celebrating Johann Sebastian Bach, with a huge
stacked headline 'Bach Is For Everyone!' in a high-contrast didone-
style display serif filling most of the canvas on pure white, a thin
hairline rule with a centered diamond beneath it, and a tiny tracked-caps 
tagline at the bottom.

Compositional deconstruction
Background: Pure white full-bleed background, edge-to-edge, no
texture or gradient. The typography itself is the entire composition
- deep ink-black type dominates against the stark white field for
maximum value contrast.

Elements
Type: text
Text: Bach Is For Everyone!
Desc: Huge stacked headline filling roughly 80% of the canvas, centered
horizontally and broken across two lines at the natural word break. High-
contrast didone-style display serif (Bodoni/Didot character) with crisp
thick-thin stroke modulation, generous x-height, and elegant
flourishes on ascenders and descenders. Deep ink-black on white,
tight leading between the two lines, letterforms nearly touching the side
margins. A tasteful italic flourish sits on the exclamation mark's dot.
Type: ob]
Desc: Small decorative flourish ornament centered just below the
headline as a classical music-program dıvider. Thin black hairline
horizontal rule spanning a short width, with a tiny solid black
diamond accent centered on the rule.
Type: text
Text: A MUSICAL MANIFESTO
Desc: Tiny tracked-caps tagline centered along the very bottom edge
of the canvas. Deep ink-black, regular-weight sans-serif or matching serif
caps with wide letterspacing, horizontal, centered alignment.

(It is a lot of text but it does describe the output pretty well. As you probably can tell I would like the option to turn the magic prompt off if I don’t want that much detail.)

Let’s create another image with some text in it. I kept the model and the aspect ratio the same, but changed the text prompt to “create the text “Classical Music is always in Style.” in a bold golden colored font centered in the image, the background should be an image of a classical era salon in an Italian palace”.

Here is a screen capture of my favorite sample image for this prompt.

a screen capture of a text poster with a photographic background showing the right sidebar

I decided to download a copy of this image to my computer so I clicked the download icon to the right of the image and as you can see a drop-down menu appeared that lets me choose the file type (choices: JPEG, PNG, or WEBP formats) and the resolution (size in pixels). The default size is 2K which Ideogram sets as 2240 pixels wide by 1792 pixels tall. Notice that you can also upscale the image from this .

Feature 2: Editable (Layerized) Text

A new feature that is still in a “beta” version is layerizing text in an AI image so that it can be edited without having to regenerate the entire image.

To test the editable text feature, I went back to the first poster image that I created above. I click the Layerize text button and Ideogram analyzes the image and finds the text so that it can be edited. It returns the image and as I mouse over the text blocks a blue outline appears.

an image that has been layerized to make the text editable

Clicking the text allows you to change the text itself and brings in a floating text toolbar that lets you change the font face and font size, justify the font, and set it to bold, italic, underlined, or strike-through.

This worked well. I changed the font face of the top line and the second line to Akaya Kanadaka (the library of open source Google fonts is included in the drop-down). I clicked the circle before the font drop-down and I got a color picker to change the color of the text.

a screen capture showing a color picker for changing text color

I choose to change the text to red (#f30b0b) and the result is shown below.

an image showing an edited text layer from an image that had been layerized

I downloaded the image and inserted into a page to see how it worked. I think it worked well — just as advertised. (I am thinking that since this is still beta — they may be still working on it.)

I was curious to see if Ideogram could edit text in an image if it wasn’t created by Ideogram. I did four tests.

The image below shows the result of one of my tests. The upper image is the original one that I created a few weeks ago. The lower image is the one I edited in Ideogram.

an image showing an original image that was not created in Ideogram (Nano Banana 2 at OpenArt) at the top and the one that was edited in Ideogram

In the image above, the top (original) image was created with Nano Banana 2 at OpenArt.ai. Ideogram did find text in the image and put the text in a blue outlined box but it didn’t find all of the text from the original. Some of the text was just removed from the background. Also I could only use 1 font face and 1 color in the text box. (I thinked I could have created a second text box in the editor and given it different properties but I didn’t go that far.)

Ideogram did let me edit text in 3 out of my 4 tests. In the fourth test it threw an error message that said it was “unable to find editable text”. So it does work but your results may vary. (Still pretty cool.)

Feature 3: JSON Prompting

I think this is a really powerful feature. I have been very interested in JSON prompt for a while. (And I have written several article about how to use it. If you aren’t familiar with JSON prompting you may want to view them. **Here is a list**.)

The Basics of JSON

What is JSON? JSON is an acronym that stands for JavaScript Object Notation. Basically it is a structure text file. The structure lets a machine (a computer) read data values out of the text so that it doesn’t have to come from a database.

Each item in the structure is an object. Objects can contain other objects so the structure can group together lots of items. Objects begin with an opening curly bracket ( { ) and ends with a closing curly bracket ( } ).

In between the opening and closing bracket is usually a name-value pair (also called a key-value pair) or multiple name-value pairs. The “name” is usually a string value (text) inside double quotes. The “value” can be another string, a number, a boolean value (true or false, 1 or 0), an array (a group of values), or another object (called a nested object).

Here is an example of a JSON object that has two name-value pairs.

{
  "name": "David",
  "age": 66
}

The first pair has a name value — “David”, and an age — 66 (no quotes because it is a number).

That is the basics of JSON prompting. Send me a response if you need more details (or **read an article**). (If JSON seems too technical you can skip ahead to the next section.

JSON Prompting for Ideogram

How does this work with Ideogram. Your prompt for Ideogram may pasted into the prompt box as a JSON prompt. Ideogram will read the JSON and create the image for you.

Ideogram 4.0 understands JSON prompting because JSON was included in the training phase of the 4.0 model. Because of this JSON prompting is rendered very quickly.

An Ideogram JSON prompt contains three primary top-level fields (think of “names” or “keys”). Only one is strictly required, though the others are highly recommended.

1. high_level_description (Optional but strongly recommended): A 1–2 sentence summary of the entire image scene.
2. style_description (Optional): Directs the aesthetics, lighting, medium, and colors.
3. compositional_deconstruction (Required): The core layout engine. It breaks down the background and the individual foreground elements.

{
  "high_level_description": "(optional) 1 or 2 sentences",
  "style_description": "(optional)",
  "compositional_deconstruction": "description of background and elements"
}

Now I will add some details to the values for the style_description and compositional_deconstruction objects.

We have nested 5 name-value pairs inside the style_description for aesthetics, lighting, medium, photo, and color_palette.

The compositional_deconstruction object has 1 name-value pair describing the background, the elements name has a value that is an array of two elements. The first element has a type of object, a bounding box, a description, and a color_palette. The second element is a text object, with a bounding box, the text to render, a description, and a color palette

Here is the expanded JSON object:

{
  "high_level_description": "A minimalist cinematic poster featuring a sleek ceramic coffee mug on a wooden table.",
  "style_description": {
    "aesthetics": "clean minimalist composition, commercial advertising style",
    "lighting": "soft, dramatic side lighting from a window with deep shadows",
    "medium": "photograph",
    "photo": "sharp focus on the subject, shallow depth of field with a blurred background",
    "color_palette": ["#1A0F0A", "#E6D7C3", "#8B5A2B"]
  },
  "compositional_deconstruction": {
    "background": "A softly blurred, dark ambient cafe background with subtle warm wood textures.",
    "elements": [
      {
        "type": "obj",
        "bbox": [350, 300, 850, 700],
        "desc": "A matte black ceramic mug filled with steaming hot coffee, sitting on a rustic dark oak table",
        "color_palette": ["#1A0F0A"]
      },
      {
        "type": "text",
        "bbox": [150, 200, 280, 800],
        "text": "MORNINGS",
        "desc": "Clean, bold white sans-serif typography centered in the upper third of the frame",
        "color_palette": ["#E6D7C3"]
      }
    ]
  }
}

Here is the JSON prompt pasted into the prompt box. The aspect ratio is set to 1:1. I click the up arrow and Ideogram begins creating the image.

a screen capture of the JSON prompt pasted into the prompt box

And here is the image created by the JSON prompt.

the image created by the JSON prompt — created by the author with Ideogram v.4.0

You may be thinking that JSON is a lot of work, however, the advantage is that once you have created the prompt you can save it and re-use it later.

I saved my prompt above, and then made a few edits to create a different image. I changed the color of the mug and the text, changed the color of the table top, and brightened the background café setting. I saved the edited file with a different file name.

the image created by the second edited JSON prompt — created by the author with Ideogram v.4.0

I made more edits — changing the mug to a wine glass of chardonnay, on a metallic table, in an elegant dining room with candlelight, I changed the font, style, and language of the text element. I save the file with another name and then pasted the edited JSON prompt into the Ideogram prompt box. Here is the image it created.

another image created by the third JSON prompt — created by the author with Ideogram v.4.0

So I was able to use a working JSON structure to create three different images and I saved all of the files so I can reuse them when I need them.

For the three images above the medium value in the style_description was set to “photograph” now I will change the medium to “pencil sketch” and edit the other values to fit a pencil drawing. I also changed the text element to “Morning Joe”. Here is the modified JSON structure. Notice the addition of an art_style value with more details.

{
  "high_level_description": "A detailed pencil sketch poster featuring a sleek ceramic coffee mug on a wooden table.",
  "style_description": {
    "aesthetics": "clean minimalist composition, monochrome fine art sketch style",
    "lighting": "soft side lighting with delicate graphite cross-hatching and deep pencil-shading shadows",
    "medium": "pencil sketch",
    "art_style": "fine graphite lines, visible paper texture, hand-drawn cross-hatching and stippling",
    "color_palette": [
      "#111111",
      "#CCCCCC",
      "#FFFFFF"
    ]
  },
  "compositional_deconstruction": {
    "background": "A softly blended, vignetted background rendered with light graphite smudging to suggest a dark ambient space.",
    "elements": [
      {
        "type": "obj",
        "bbox": [
          350,
          300,
          850,
          700
        ],
        "desc": "A hand-drawn ceramic mug shaded with dark charcoal pencil tones to appear matte black, resting on a sketch of a rustic wooden table",
        "color_palette": [
          "#111111"
        ]
      },
      {
        "type": "text",
        "bbox": [
          150,
          200,
          280,
          800
        ],
        "text": "Morning Joe",
        "desc": "Clean, bold negative-space typography cleanly erased out of the background shading, centered in the upper third of the frame",
        "color_palette": [
          "#FFFFFF"
        ]
      }
    ]
  }
}

Here is the image.

a new image created by the JSON prompt with a different medium value — created by the author with Ideogram v.4.0

I would like to change the medium one more time. This time I will change it to a “crayon drawing” with a text element of “Dad’s Coffee”.

Here is the edited JSON.

{
  "high_level_description": "A vibrant and colorful child-like crayon drawing poster featuring a coffee mug on a table.",
  "style_description": {
    "aesthetics": "naive art style, playful childlike composition, energetic and fun",
    "lighting": "bright, even lighting with minimal flat shading drawn with heavy hand pressure",
    "medium": "crayon drawing",
    "art_style": "waxy crayon texture, visible heavy wax strokes, raw paper texture showing through loose coloring",
    "color_palette": [
      "#FF0055",
      "#FFCC00",
      "#33CCFF",
      "#000000"
    ]
  },
  "compositional_deconstruction": {
    "background": "A background filled in with scribbled electric blue crayon strokes, creating a bright energetic space.",
    "elements": [
      {
        "type": "obj",
        "bbox": [
          350,
          300,
          850,
          700
        ],
        "desc": "A thick, heavily-colored bright pink wax crayon mug with stylized white loops for steam, sitting on a sun-yellow crayon table surface",
        "color_palette": [
          "#FF0055",
          "#FFCC00"
        ]
      },
      {
        "type": "text",
        "bbox": [
          150,
          200,
          280,
          800
        ],
        "text": "Dad's Coffee",
        "desc": "Bold typography hand-drawn with a thick black wax crayon block letters, centered in the upper third of the frame",
        "color_palette": [
          "#000000"
        ]
      }
    ]
  }
}

Here is the image.

another new image created by the JSON prompt with another different medium value — created by the author with Ideogram v.4.0

I think I would have to rework the text descriptions a bit — this looks to me more like it was drawn with colored markers but it still shows how different an image can be with an edited JSON file.

Feature 4: Bounding Boxes

This is actually part of the JSON prompting feature but it is innovative nonetheless.

You may have noticed in the JSON displayed in the previous section that many of the elements had a bbox value that looked like this.

{
  "elements": [
    {
      "type": "text",
      "bbox": [
        150,
        200,
        280,
        800
      ],
      "text": "Dad's Coffee",
      "desc": "Bold typography hand-drawn with a thick black wax crayon block letters, centered in the upper third of the frame",
      "color_palette": [
        "#000000"
      ]
    }
  ]
}

The bbox value is an array of 4 numeric values. The array begins with “ [ “ and ends with “ ] “. The four numbers represent the coordinates of the top left corner, and bottom right corners of the bounding box. The coordinates 0,0 represent the top corner of the image itself. So, the text “Dad’s Coffee” will be located in a box that begins 150 units down from the top edge and 200 units in from the left edge of the image and continue to 280 units down from the top edge and 800 units in from the from edge. (This would outline a rectangular area 130 units tall and 600 units wide.)

This allows the prompter to precisely position text elements on “canvas” of the image.

Remember that elements can also be objects. so an item (like a coffee mug or a person) from a reference image could be precisely positioned in the image as well.

The bounding box feature of the JSON prompting allows the prompter place many elements in the image for compositional layout of the items.

Feature 5: Transparent Background

The training of the Ideogram v.4.0 model also included training on images with transparent backgrounds.

It is easy to create your images with a transparent background by including the words “transparent background” in your prompt.

Here is an image of a raccoon created with a transparent background.

a screen capture of the prompt and the raccoon image on a transparent background — the dotted background is transparent — created by the author in Ideogram v.4.0

Let’s try generating an image with 3 separate characters with a transparent background to see the clean edges of the transparency between the characters. Let’s make a cartoon of the 3 Little Pigs.

a screen capture of the prompt box for the 3 Little Pigs image with a transparent background

Here is the image.

the 3 Little Pigs image — the dotted background is transparent — created by the author in Ideogram v.4.0

Let’s combine this with a cartoon background image in PhotoPea. Super Easy!

a screen capture of the 3 little pigs image with transparency dropped into another background image in PhotoPea

Ideogram also has a background removal tool to remove backgrounds from images that have already have been created. It works really well!

I think that the new and enhanced features of Ideogram 4 are great, and the fact that they are providing the inference code and the weights as open source on Github is mind blowing.

I like all of the new features. The only thing that I don’t like is that the magic prompt option is on by default is v.4.0. I would like for them to make it optional.

Thank you for reading my article and supporting my writing.

I enjoy learning and sharing information about AI Image Generators. I have been using Ideogram for a few years now and I really like to use it. I have always felt that it created good images, but this latest model (v.4.0) is quite extraordinary.

Let me know what you think of my articles and if you liked this one. Send me your comments, criticisms, and suggestions of what I can do better.

You can find all of my articles on **my Medium profile**.

Thanks again for reading and sharing, and have a great week! David

[embed]Midjourney Images Inspired By Medium Writers Finding Inspiration to Create AI Images from Articles on Mediummedium.com

[embed]My Midjourney Info 17 — Week of June 1st, 2026 Midjourney News and Office Hours Recap, 14 Blended and 35 Individual SREF IDs, and 20 Moodboardsmedium.com

[embed]Midjourney Artistry Shared on Facebook — Introducing Another Artist I Have Learned a lot from Following This Artist and Her Postsmedium.com

**A List of My JSON Prompting Articles online **— Easy Access to lots of Information about this type of Prompting.

**A List of My Articles about Style References at Midlibrary** — view a lot of Midjourney SREFs.