Can you use two characters in one Ideogram AI image?
Yes — you can put two or more characters in one Ideogram image by describing both in your prompt. But as of 2026-07 you can lock only one of them to a Character Reference: the API accepts exactly 1 reference image (10MB max, JPEG/PNG/WebP). To pin both faces, generate the scene, then Magic Fill each face separately.
Why — the first-principles explanation
Two different questions hide inside this one, and they have opposite answers. Can Ideogram draw two people in a single image? Of course. Write "two women laughing at a kitchen table" and you get two women. Nothing limits how many people a prompt can describe. Can Ideogram make two specific people — two faces you uploaded — appear in the same image? As of 2026-07, no, not in a single pass. Ideogram's official API documentation states the character reference parameter "currently only supports 1 character reference image."
The reason is a binding problem, not laziness. When you upload a reference photo, Ideogram detects the face and hair and compresses them into what amounts to an identity map — a compact numerical description of what makes that person look like themselves. The generator then pulls every face it draws toward that one map. With a single identity slot, the system has no way to answer the question "which face goes where?" If you handed it two maps, it would need a second mechanism to route identity A to the person on the left and identity B to the person on the right. Without that routing, both faces drift toward an average of the two people, or one identity wins and you get accidental twins.
This is also why Ideogram's own best-practice guidance tells you to avoid reference photos containing multiple people. A group photo gives the face detector several candidates and no rule for picking one. The model may lock onto the wrong face, or blend two of them. One clear, well-lit, front-facing photo of one person is the input the system is actually built for.
The workaround exploits the same logic. Magic Fill lets you paint a mask over one region of an existing image and regenerate only that region with a character reference applied. The mask supplies the missing routing information — you are pointing at the pixels and saying "this face, right here." So you generate the two-person scene first, then run two separate Magic Fill passes, one face at a time, each with a different reference. One identity slot is enough when you use it twice.
An example that makes it click
Think of a passport-photo booth that can only hold one photo in its scanner at a time. You want a picture of your two kids sitting on a bench together. The booth can happily draw a bench with two kids on it — that part is just a description. But when you slide in a photo of your daughter and ask it to make one kid look like her, the machine has no way to know which kid you mean. So it makes both of them look a bit like her. You get sisters who aren't quite sisters.
The fix is to point. First, get the bench-and-two-kids picture. Then cover everything with paper except the face on the left, slide in your daughter's photo, and let the booth redo just that hole. Then cover everything except the face on the right, slide in your son's photo, redo that hole. Two passes, two photos, one machine — and now both kids are the right kids. The paper mask is doing the job your finger would do if you could just say "that one."
How to do it
- Write a prompt that describes both characters and the scene, with clear positions — for example, "two women at a kitchen table, one on the left in a red sweater, one on the right in a denim jacket." Generate until the composition and posing look right.
- Ignore facial likeness at this stage. You are only building a base image with two well-placed, unobstructed, front-facing or 3/4-view faces.
- Open the winning image in Ideogram Canvas and select Magic Fill from the left sidebar.
- Brush a mask over the first character's face and hair only. Click Next.
- Click Character in the prompt box and select or upload your first reference image — one person per photo, well-lit, minimal filters. Generate.
- Repeat Magic Fill on the second character's face with your second reference image. Adjust the mask if you want to keep a hat, glasses, or a hairstyle from the base image.
- Inspect the seams. If lighting or skin tone mismatches between the two passes, re-run the weaker pass with a slightly larger mask so it blends more surrounding pixels.
Key facts
- Ideogram's official API documentation for Ideogram 3.0 states the character reference parameter "currently only supports 1 character reference image" (as of 2026-07).
- Character reference images have a maximum total size of 10MB across all references, and must be JPEG, PNG, or WebP.
- There is no limit on how many people a text prompt can describe — the 1-image cap applies only to identity-locked characters, not to bodies in the frame.
- Ideogram Character, the single-image character-consistency model, launched July 29, 2025, and requires no LoRA training or multi-photo dataset.
- Official best-practice guidance says to avoid reference photos with multiple characters, blur, or heavy distortion; frontal or 3/4 views work best.
- When a character reference is applied, Ideogram automatically masks the face and hair from the reference image; the mask can be manually adjusted.
▶ The 60-second explainer (script)
Can you use two characters in one Ideogram image? Yes — but there's a catch worth understanding. If you just want two people in the picture, describe them in your prompt. Two women at a table, a knight and a dragon, whatever you like. Ideogram draws as many people as you ask for. That part was never limited. The catch is character references — the feature that locks a face to a specific person you uploaded. As of July 2026, Ideogram's own API docs say it supports exactly one character reference image, with a ten-megabyte cap, in JPEG, PNG, or WebP. Why only one? Because with a single identity slot, the model has no way to know which face in your scene should be which person. Hand it two references and it can't route them — you get two people who both look like a blend of your uploads. So here's the fix. Generate your two-person scene first and don't worry about likeness. Then open Magic Fill, brush a mask over just the first face, and apply your first character reference. Then do it again on the second face with your second reference. The mask tells the model exactly which face you mean. Two passes, two references, one image. One more tip: your reference photos should each contain only one person. A group photo gives the face detector too many choices and it will pick wrong.
What authoritative sources say
People also ask
Can I upload one photo that already has both people in it?
You can, but it works badly. Ideogram's guidance is to avoid reference photos with multiple characters, because the face detector has no rule for choosing which person to lock onto. It may pick the wrong face or blend the two.
How many people can appear in one Ideogram image overall?
There is no fixed cap — that is governed by your prompt, not by the reference limit. Likeness quality drops as faces get smaller and more numerous, so crowds render as generic people rather than specific ones.
Will Ideogram ever support two character references at once?
The API documentation says "currently only supports 1," which signals the limit is a present-state constraint rather than a permanent design choice. Check the official developer docs before assuming it still applies.
Does the Magic Fill workaround cost more?
Each pass is a separate generation, so two faces means at least two billed operations on top of your base image. Generations using character references are priced differently from standard generations as of 2026-07.
Why do my two Magic Fill passes look mismatched?
Each pass regenerates its region independently, so lighting and skin tone can drift. Widening the mask slightly gives the model more surrounding pixels to match against, which usually blends the seam.