Sign in
Workflow guide

How to create a consistent character with AI in Ideogram

To keep the same character across many images in Ideogram, attach one to five reference images of that character to Ideogram 4.5 and describe the new scene. In the app, each image costs 6 credits at Medium quality. Through the API, the same request costs $0.06, and Ideogram 3.0 has separate Character endpoints that take one reference image at $0.10 to $0.20 per image.

This guide puts the models and limits next to the prices, then tests one fictional character across six scenes and reports what held and what drifted. Every prompt appears as it was run.

Prices and limits were checked on November 16, 2026.

In this guide

Overview : Consistent characters in Ideogram at a glance

The model, the file rules, the cost and the test result for consistent characters in Ideogram
Question Answer
Model in the app Ideogram 4.5, with one to five reference images
Model in the API Ideogram 4.5 generate with up to five images, or Ideogram 3.0 Character with one character_reference_images file
Reference files JPEG, PNG or WebP, up to 25 MB each
Ideogram 4.5 with references 0.8 to 22 credits per image in the app, or $0.008 to $0.22 in the API, by quality
Ideogram 3.0 with a character reference, API $0.10 at Turbo, $0.15 at Default, $0.20 at Quality
Result of the test in this guide From one reference, the character stayed recognizable in all six scenes. Clothing the reference did not show changed from scene to scene
Reference The reference portrait, cropped to the face: copper curls with a white streak, freckles, round gold glasses and one hoop earring.
Scene 1 Her face in the cafe scene, turned to one side in dusk light.
Scene 2 Her face in the snowy platform scene, above a cream scarf.
Scene 3 Her head in the harbor side view, seen in profile against a blue sky, with no earring on the visible ear.
Scene 4 Her face in the leap scene, laughing with her mouth open.
Scene 5 Her face in the gouache illustration, with the white streak and round glasses redrawn.
Scene 6 Her face in the market scene, looking at the camera.
These crops come from the reference image and the six scenes Ideogram 4.5 generated from it. The glasses, freckles and copper curls carry across. The white streak changes width from scene to scene, and the illustrated version lightens her skin.

Overview : Which Ideogram model keeps a character consistent in the app and in the API?

Ideogram 4.5 is the model for character consistency in the Ideogram app, and it accepts the same reference images through the API. Ideogram 3.0 offers a second route: the Ideogram 3.0 Character app for a saved character, and a set of Character endpoints in the API.

The four ways to keep a character consistent in Ideogram, by model and by where each one runs
Where Model What you give it How you reach it
App, reference images Ideogram 4.5One to five images, attached to each runImage app, Reference images field
App, saved character Ideogram 3.0 CharacterOne image, saved once with a nameLibrary, then Characters
API, source images Ideogram 4.5Up to five files in imagesPOST /v2/image/generate/ideogram-4-5
API, character reference Ideogram 3.0One file in character_reference_imagesPOST /v2/image/generate/ideogram-3-character

The two routes are separate. Per Ideogram’s saved characters guide, a saved character works in the Ideogram 3.0 Character app, where your workspace has it, and Ideogram 4.5 uses the images you attach to each run. Ideogram 4.0 and Ideogram 3.0 take no reference images in the Image app.

Style presets, style reference images, negative prompts and color palettes belong to Ideogram 3.0, as the Styles page explains. Ideogram 4.5 takes a look from the prompt or from another reference image.

Overview : How many reference images does Ideogram accept, and what makes a good one?

Ideogram 4.5 accepts one to five reference images, and each file must be JPEG, PNG or WebP, up to 25 MB. The Ideogram 3.0 Character app and the Ideogram 3.0 Character API take one image.

The API enforces both limits. In dry runs on November 16, 2026, a sixth file in images was rejected with a 400 error, and so was a second file in character_reference_images.

Ideogram’s reference images guide asks for clear images of the character and says shots of the face from a few angles work best. For a saved character, the guidance is one centered, front-facing image against a simple background.

The test below adds one rule: the reference decides what the model has to invent. The reference in this guide stops at the chest, so the scenes came back with different trousers and shoes.

Studio portrait of a fictional woman with short copper curls and a white streak, round gold glasses, a mustard jacket, a striped shirt and a teal paper-boat pin.
The reference image was generated with Ideogram 4.5 for this guide. The character is fictional. The prompt asked for a waist-up portrait and the result is closer, so nothing below the chest is defined.

Prompt as run

A studio portrait photograph of a woman in her early thirties, shown from the waist up, facing the camera with a slight closed-mouth smile. She has warm brown skin, a round face, freckles across her nose and cheeks, grey-green eyes, and short, tightly curled copper-red hair with one white streak at the front above her left temple. She wears round gold wire-rim glasses and one small gold hoop earring in her left ear. She wears a mustard-yellow cotton chore jacket over a navy and white striped shirt, with a small teal enamel pin shaped like a paper boat on the left lapel. Plain light grey seamless background, soft even daylight, sharp focus, natural skin texture.

The settings were size 1728x2304, quality medium, Magic Prompt off and seed 4521.

How to : How do you create a consistent character in the Ideogram app?

In the Ideogram app, you create a consistent character by choosing Ideogram 4.5, adding reference images of the character and describing each new scene. These steps follow Ideogram’s reference images guide.

  1. Open the Image app and choose Ideogram 4.5 in the Model picker.
  2. Add one to five clear images of the character in Reference images.
  3. Type @ in the prompt, pick the character’s image, then describe the scene. Ideogram’s example is @img_1 sitting at a café table, reading a newspaper, morning light.
  4. Click Generate. The button shows the cost in credits before you run it.
  5. Attach the same images each time you want that character again, and compare every result with them.
The Ideogram 4.5 form with one reference image attached, tagged @img_1, and a prompt that reads “Place the woman from @” with a menu offering img_1.
In the Ideogram app, attach up to five reference images, then type @ in the prompt to point at one. Each image gets a tag such as @img_1.

Once a reference is attached, the prompt works as an edit instruction, Magic Prompt disappears and Quality gains a Very low option. If you switch to a model that takes no reference images, Ideogram clears them.

If your workspace has the Ideogram 3.0 Character app, you can save a character instead. Open Library, then Characters, select Create character, name it and attach one image. Select Use on its card to generate with it.

How to : How do you keep a character consistent with the Ideogram API?

With the Ideogram API, you keep a character consistent by sending its images in the images field of the Ideogram 4.5 generate endpoint and writing the prompt as an instruction. The first image is the one being edited, and any others are extra references.

This is the request behind the first scene in the test:

Shell
curl -X POST https://api.ideogram.ai/v2/image/generate/ideogram-4-5 \
  -H "Api-Key: $IDEOGRAM_API_KEY" \
  -F [email protected] \
  -F prompt="Place the woman from this image in a new scene. She sits at a small round table outside a cafe on a rainy city street at dusk, both hands around a white ceramic cup, looking out at the street. Medium shot, warm light from the cafe window. Keep her face, freckles, hair, glasses, earring, jacket, striped shirt and lapel pin the same." \
  -F size=1728x2304 \
  -F quality=medium \
  -F magic_prompt=off \
  -F seed=4521

For several references, repeat -F [email protected] for each file. Add ?dry_run=true to the URL to get a price quote with nothing generated or billed.

The request fields that carry a character on the Ideogram 4.5 generate endpoint and on the Ideogram 3.0 Character generate endpoint
Setting Ideogram 4.5 generate Ideogram 3.0 Character generate
Character input images, up to 5 files of 25 MB eachcharacter_reference_images, one file of up to 25 MB
Mask mask limits the edit to part of the first imagecharacter_reference_mask marks where the character is in the reference
Quality or speed quality: very_low, low, medium or highrendering_speed: turbo, default or quality
Style controls The prompt, or a reference imagestyle_type, style_reference_images (up to 10) and negative_prompt

The Ideogram 3.0 Character API has two more endpoints. Remix restages a source image with your character, and inpaint repaints a masked area of a scene so it features your character. The Ideogram 4.5 generate reference and the Character generate reference list every parameter.

How to : How much does a consistent character cost in credits and dollars?

A character image from Ideogram 4.5 with reference images costs 6 credits in the app, or $0.06 through the API, at Medium quality. One credit is one US cent, so the two columns below match.

The cost per image of a consistent character in app credits and in US dollars through the API, by model and setting
Model and setting App credits per image API price per image
Ideogram 4.5 with references, Very low 0.8$0.008
Ideogram 4.5 with references, Low 3$0.03
Ideogram 4.5 with references, Medium 6$0.06
Ideogram 4.5 with references, High 22$0.22
Ideogram 3.0 with a character reference, at Turbo, Default or Quality Shown on the Generate button$0.10, $0.15 or $0.20

Two details matter when you budget, both checked on November 16, 2026:

  • The price does not change with the number of reference images. A dry run with three references at High quoted the same $0.22 as one.
  • High with references costs $0.22, against $0.10 for a prompt-only image, so Medium is the cheaper setting to iterate at.

Reference-image generations use credits on every plan and need priority credits. On the pricing page, Plus includes 2,400 priority credits a month, which is 400 character images at Medium. The API pricing tab lists the live rates, and the Ideogram API pricing guide works through larger budgets. The 23 API calls behind this guide cost $1.46.

Very low, $0.008
A wider view of the same woman at a cafe table beside a lit window at dusk, generated at Very low quality.
Medium, $0.06
The same fictional woman at a cafe table on a rainy street at dusk, holding a white cup.
The same prompt and seed produced these two images at Very low for $0.008 and at Medium for $0.06. The character is recognizable in both. At Very low the composition differs, the white streak spreads across the hairline, and a sign in the background has garbled lettering.

The test : How well does one reference image hold across six scenes?

From one reference image, Ideogram 4.5 kept the character recognizable in all six scenes of this test. The glasses and copper hair appear in every scene, and the face matches in the four photographic scenes that show it from the front. Most of what changed was something the reference did not show.

Each scene was one API call with the reference attached, at size 1728x2304, quality medium, Magic Prompt off and seed 4521. The images below are the first result of each prompt. Nothing was rerun or retouched.

The same fictional woman at a cafe table on a rainy street at dusk, holding a white cup.
Scene 1 is a new setting. Notice the pin, the single earring and the striped shirt, all carried over.

Prompt as run

Place the woman from this image in a new scene. She sits at a small round table outside a cafe on a rainy city street at dusk, both hands around a white ceramic cup, looking out at the street. Medium shot, warm light from the cafe window. Keep her face, freckles, hair, glasses, earring, jacket, striped shirt and lapel pin the same.

The same fictional woman on a snowy train platform in a dark green overcoat and a cream scarf.
Scene 2 is a different outfit. Notice that the face survives without the signature jacket, and that a second thin ring has appeared beside her ear.

Prompt as run

Place the woman from this image in a new scene and change her outfit. She stands on a snowy train platform in pale morning light, wearing a dark green wool overcoat and a cream knitted scarf, hands in her pockets, looking at the camera. Medium shot. Keep her face, freckles, hair, glasses and earring the same.

The same fictional woman in side view, pushing a blue bicycle along a harbor wall.
Scene 3 is a different angle and distance. Notice the shorter jacket and the invented trousers and shoes.

Prompt as run

Place the woman from this image in a new scene. Wide shot, full body, seen from the side in profile. She walks along a stone harbour wall at midday, pushing a blue bicycle, fishing boats in the water behind her. Keep her face, freckles, hair, glasses, earring, jacket, striped shirt and lapel pin the same.

The same fictional woman leaping over a puddle on a city street, laughing.
Scene 4 is an action pose. Notice the jeans and white trainers, which appear in no other scene.

Prompt as run

Place the woman from this image in a new scene. She leaps over a wide puddle on a city pavement, caught in mid-air with both feet off the ground and arms out for balance, laughing with her mouth open. Full body, bright overcast daylight, slight motion blur in the background. Keep her face, freckles, hair, glasses, earring, jacket, striped shirt and lapel pin the same.

A gouache picture-book illustration of the same character reading a map by lantern light in a tent.
Scene 5 is a different art style and lighting. Notice which markers survive the redraw: the streak, the glasses, the earring and the pin.

Prompt as run

Redraw the woman from this image as a children's picture-book illustration in gouache, with flat shapes and visible brush texture. She sits cross-legged inside a canvas tent at night, reading a paper map by the light of a lantern. Keep her freckles, copper curls with the white streak, round glasses, earring, mustard jacket, striped shirt and teal paper-boat pin recognisable in the new style.

The same fictional woman at a market stall holding a wooden sign that reads MAPS FOR PEOPLE WHO GET LOST, Open 9 to 5, Tuesday to Sunday.
Scene 6 is a sign with exact text. Notice that both lines match the prompt letter for letter.

Prompt as run

Place the woman from this image in a new scene. She stands behind a market stall of rolled paper maps and holds up a hand-painted wooden sign with both hands. The sign reads "MAPS FOR PEOPLE WHO GET LOST" in large white capital letters, and below it "Open 9 to 5, Tuesday to Sunday" in smaller cream letters. Medium shot, daylight. Keep her face, freckles, hair, glasses, earring, jacket, striped shirt and lapel pin the same.

What held and what drifted in each of the six scenes generated from one reference image
Scene What held What drifted
1. Cafe at dusk Face, freckles, curls and streak, glasses, earring, jacket, shirt, pinHer gray-green eyes read as dark brown in the dusk light
2. Coat and scarf Face, freckles, hair, glasses, eye colorA second thin ring beside the hoop earring
3. Side view, wide Glasses, hair color, jacket color. The far ear has no earring, which matches the referenceA shorter jacket, invented trousers and shoes, and a streak that is almost hidden
4. Leap Face in an open laugh, streak, glasses, earring, pin, jacket, shirtInvented jeans and white trainers, and an extra label on the chest pocket
5. Gouache illustration Streak, glasses, freckles, earring, jacket, shirt and pin, all redrawnLighter, pinker skin, curls drawn as flat waves, and a looser likeness
6. Sign All 13 words on the sign, face, glasses, earring, pin, jacketThe streak is wider and the hair a little lighter

Can the character hold a product with exact label text?

A second reference image can carry a product. This test generated a fictional tea tin, then sent the character as the first image and the tin as the second.

Product reference
A teal tea tin with a brass lid and a cream label that reads HARBOUR FOG, Smoked black tea, 100 g, under a line drawing of a lighthouse.
Result
The same fictional woman in a kitchen, smiling and holding the Harbour Fog tea tin beside her face.
Label detail
A close crop of the tin in the result: the label reads HARBOUR FOG, Smoked black tea, 100 g, under the lighthouse drawing.
The product reference is shown cropped at the sides, next to the result and a detail of its label. All three lines of the label and the lighthouse drawing match. The tin’s proportions changed: the lid is taller and the teal band above the label is thinner.

Prompt as run

Place the woman from the first image in a bright kitchen by a window. She holds the tea tin from the second image up beside her face with one hand, label facing the camera, and smiles. Keep the tin's shape, colours and label text exactly as in the second image. Keep her face, freckles, hair, glasses, earring, jacket, striped shirt and lapel pin the same. Medium shot, morning light.

The test : Do two or three reference images work better than one?

In this test on Ideogram 4.5, a second reference image fixed the character’s clothing and left the face as it was. The face already held with one reference, so the extra images helped by showing what the first one lacked.

The two extra references were made from the first one for $0.06 each: a full-length view that defines the jacket, trousers and boots, and a left profile.

Portrait
Studio portrait of a fictional woman with short copper curls and a white streak, round gold glasses, a mustard jacket, a striped shirt and a teal paper-boat pin.
Full length
Full-length studio view of the same fictional woman in a mustard chore jacket, navy trousers and brown boots.
Left profile
Left profile of the same fictional woman against a gray backdrop, with the hoop earring visible.
This is the reference set. The full-length view was generated from the portrait with a prompt that named navy trousers and brown lace-up boots. The profile came from the same portrait, and its pin lost the paper-boat shape.
One reference
The same fictional woman leaping over a puddle on a city street, laughing.
Two references
The same woman jumping over a puddle, wearing navy trousers and brown boots.
Three references
The same woman in a star jump over a puddle, wearing navy trousers and brown boots.
The leap was run with one, two and three references, with the same prompt and seed. With the full-length reference added, the trousers, boots and jacket length match it.
One reference
The same fictional woman in side view, pushing a blue bicycle along a harbor wall.
Outfit described
The same woman in side view with a blue bicycle beside a stone wall, wearing a hip-length mustard jacket, navy trousers and brown boots.
Three references
The same woman in side view with a blue bicycle on a harbor wall, wearing a hip-length mustard jacket, navy trousers and brown boots.
Her head in the harbor side view, seen in profile against a blue sky, with no earring on the visible ear.
Her head in the side view with the outfit described, with no earring on the visible ear.
Her head in the three-reference side view, with a hoop earring on the visible ear.
The side view was run with one reference, with one reference plus a sentence describing the outfit, and with three references. The lower row enlarges each head. Both fixes produce the right outfit. In the three-reference version the hoop earring is on the wrong ear.
How each reference setup affected the outfit, the face and the time per call in this test
Setup Outfit matches the full-length reference Face Time per call
One reference No. It changed in each sceneHeld21 to 26 seconds
One reference, outfit described in the prompt YesHeld24 seconds
Two references: portrait and full length YesHeld33 seconds
Three references: portrait, full length, profile YesHeld, with the earring on the wrong ear in the side view34 to 38 seconds

Every call in this table cost $0.06.

The test : What does the Ideogram 3.0 Character API return for the same scenes?

On the same reference image, the Ideogram 3.0 Character endpoint kept the face and hair, and the clothing followed the prompt. The test ran two calls, one at Turbo for $0.10 and one at Default for $0.15.

Turbo
An Ideogram 3.0 Character result: a woman with copper hair and a white streak holding a cup at a cafe table, in a mustard shirt jacket with a large paper-boat brooch.
Default
An Ideogram 3.0 Character result: a woman with copper curls holding a dark sign that reads MAPS FOR PEPLE WHO GET LOST, OPEN 5 TUESDAY TO SUNDAY.
These Ideogram 3.0 Character results were generated at Turbo and at Default, 864x1152 each, with style_type realistic and Magic Prompt off. The face, freckles and streak match the reference. The jacket is a different garment.

Prompt as run at Turbo

A photograph of a woman sitting at a small round table outside a cafe on a rainy city street at dusk, both hands around a white ceramic cup, looking out at the street. She wears round gold wire-rim glasses, a mustard-yellow cotton chore jacket over a navy and white striped shirt, and a small teal paper-boat pin on her lapel. Medium shot, warm light from the cafe window.

Prompt as run at Default

A photograph of a woman standing behind a market stall of rolled paper maps, holding up a hand-painted wooden sign with both hands. The sign reads "MAPS FOR PEOPLE WHO GET LOST" in large white capital letters, and below it "Open 9 to 5, Tuesday to Sunday" in smaller cream letters. She wears round gold wire-rim glasses, a small gold hoop earring in her left ear, a mustard-yellow cotton chore jacket over a navy and white striped shirt, and a small teal enamel paper-boat pin on her left lapel. Medium shot, daylight.

Both prompts described the jacket, shirt and pin, and the results followed those words rather than the reference: the jacket came back as a different cut and the pin as a large folded-paper boat. The sign read “MAPS FOR PEPLE WHO GET LOST” and “OPEN 5 TUESDAY TO SUNDAY”, where Ideogram 4.5 returned both lines exactly. Neither call sent the optional character_reference_mask.

Two calls are a small sample. They suggest using Ideogram 4.5 when clothing and lettering must match, and Ideogram 3.0 Character when you need a negative prompt or style reference images.

The test : How do you edit an image and keep the character?

To change one thing in an image that already shows your character, use Precise Edit on Ideogram 4.5. It returns the image at its original size and accepts up to four reference images, or three when you add a mask.

Before
The same fictional woman at a cafe table on a rainy street at dusk, holding a white cup.
After
The cafe scene after the edit: the same woman at the same table, now in a green jacket with no pin.
The face is unchanged to the eye. The prompt asked for a wool overcoat and the result kept the jacket’s cut in green. The pin went with the jacket, and the hem below the table stayed mustard.

Prompt as run

Change her mustard jacket to a dark green wool overcoat. Keep her face, hair, glasses, earring, striped shirt, the cup, the table and the street exactly the same.

In this test, one unmasked Precise Edit at Medium changed the jacket and left 64% of the pixels identical to the source. In the face and hair area the average change was 5 levels out of 255. The prompt did not list the pin, and it disappeared, so name everything that must stay, or mask the area as Ideogram’s masks and inpainting guide describes.

Reference : What prompt pattern keeps a character consistent?

The prompt pattern used for every Ideogram 4.5 scene in this test names the character through the reference, describes only the change, and ends with a list of what to keep.

Prompt pattern

Place the [character] from this image in a new scene. [What they do, where, in what light.] [Shot size and angle.] [Clothing the reference does not show.] Keep [their face and each item that must not change] the same.

Prompt tips for consistent characters and the source of each one
Tip Source
In the app, point to the character with its @ tag. In the API, a phrase such as “the woman from this image” did the same job Ideogram’s docs for the tag, this test for the phrase
Say what should change and keep the rest short Ideogram’s docs
End with a keep list. The prompt the API returned had sorted that list into its own section From this test
Describe clothing the reference does not show, or add a reference that shows it From this test
Put exact words in quotation marks and say where they sit Ideogram’s docs

No scene prompt in this test described the character’s face again. The reference carried it.

Reference : What are the limits and rules for character references?

Character references in Ideogram must be JPEG, PNG or WebP files of up to 25 MB, and a photo of a real person needs that person’s consent.

  • Real people. The Usage Policy restricts content that is invasive of privacy or publicity rights, and it gives uploading images of individuals without their consent as an example. The character in this guide is fictional.
  • Presentation. The same policy restricts presenting an output as an actual photograph of a real event.
  • Two characters. Give each one its own reference. The character consistency page says that keeping two identities separate is not guaranteed.

The Terms of Service hold the full terms, and the character consistency page lists more reference setups.

Reference : What goes wrong with character consistency, and how do you fix it?

Most character consistency failures in this test came from details the reference image did not define. Each fix is marked as tested here or as Ideogram’s guidance.

Character consistency failures, the fix for each and whether the fix was tested here
What went wrong Fix Status
Trousers, shoes or jacket length change between scenes Add a full-length reference, or describe the outfit in one sentenceTested here. Both worked
Misspelled words on a sign from Ideogram 3.0 Character Generate the lettering with Ideogram 4.5, or correct it with editable text layersIdeogram 4.5 tested here
The face drifts toward a generic face Add clear references that show the face from a few anglesIdeogram’s guidance

Reference : Frequently asked questions about consistent characters in Ideogram

What is character consistency in AI image generation?
Character consistency means that one character keeps the same face, features and identity across many images. In Ideogram, you get it by attaching reference images of the character to Ideogram 4.5 and describing a new pose, outfit or scene. Reusing the same references guides each later generation.
Can you keep a character consistent in Ideogram for free?
You can keep a character consistent on Ideogram’s free plan, within its weekly credits. Every plan can use Ideogram 4.5 in the Image app, and eligible free accounts receive weekly credits. Generations with reference images always use credits. Images made on the free plan are public, as the Ideogram FAQ explains.
Can you combine a character reference with a style reference?
The Ideogram 3.0 Character endpoint accepts one character reference image together with up to 10 style reference images, and dry runs on November 16, 2026 quoted the usual $0.10 to $0.20. The inpaint reference says combining a style with a character must be enabled for an API account, so check with a dry run. On Ideogram 4.5, add a style image as another reference.
When did Ideogram Character launch, and what can it do?
Ideogram’s announcement of Ideogram Character is dated July 29, 2025. It introduced character consistency from one reference image, with a mask that sets what the reference preserves. Today the Ideogram 3.0 Character API generates, remixes and inpaints with one character reference, and Ideogram 4.5 handles reference images in the app.
Can you use a photo of a real person as a character reference?
You can use a photo of a real person as a character reference only with that person’s consent. Ideogram’s Usage Policy restricts content that is invasive of privacy or publicity rights and names uploading images of individuals without their consent as an example. Start with a photo of yourself, or of someone who has agreed.
Can you make a comic or children’s book with one consistent character?
For a comic or children’s book, Ideogram’s character consistency page recommends attaching the same references for every page and changing only the scene and action. Keep the clothing description identical in each prompt and proofread any lettering. In this test, a gouache picture-book version kept the character’s streak, glasses and pin.

Start with one clear image of your character

Open Ideogram 4.5, attach one clear image of your character and describe the first scene. Add a full-length reference when the outfit has to stay the same.