Ideogram Character pins identity from a single reference photo and writes that same person into any scene, outfit, style, or genre you describe. The fastest way to build a character set without training a LoRA.
Ideogram Character solves the persistent-identity problem without training. Drop in one reference photo of a person and Ideogram pins facial features, hair, and proportions — then writes that same character into any new prompt you give it. New outfit, new scene, new genre, same face.
Three speed/quality tiers (Turbo / Default / Quality), three style modes (Auto / Fiction / Realistic), seven aspect ratio presets, and an optional magic-prompt rewrite. The result is a character set you can build in minutes — no LoRA training, no checkpoint juggling, no consistency drift.
1
Reference image needed
3
Style modes
7
Aspect ratio presets
Ideogram Character keeps the reference subject's face, hair, proportions, and key features consistent across every output — even when the rest of the scene changes radically. Different style, different lighting, different era; same person.
Style modes (Realistic, Fiction, Auto) let you push the same character across photoreal portraits, fantasy/sci-fi illustration, and stylized media. Anime variations, noir compositions, period photography — identity holds across the genre jump.
Turbo for rapid iteration — get 10 variations cheaply, pick the best. Default is the sweet spot for production work. Quality for hero shots and final renders where identity fidelity matters most. Same prompt language across all three.
Train a LoRA only when you need very specific aesthetic control or extreme fidelity. For most character-set needs — comic panels, storyboards, social campaigns, product mascots — Ideogram Character ships consistent variations in minutes instead of hours.
Each example shows the exact prompt that produced the result. Copy any prompt with one click.
Studio portrait, neutral backdrop
Studio portrait of a 28-year-old woman with short auburn pixie cut, freckles, hazel eyes, wearing a forest-green linen shirt, soft window light, neutral light-gray backdrop, three-quarter view, 85mm lens
Reference image best practice: clean lighting, neutral backdrop, three-quarter or front view, sharp eyes, no occlusion of features. A clear reference makes downstream variations dramatically more consistent.
Default speed · 3:4 · Fiction style
the same woman wearing ornate fantasy elven armor with engraved silver pauldrons, standing in a misty forest at dusk, cinematic key light from the right, soft volumetric fog, detailed atmospheric lighting
Lead with "the same woman" or "the same character" — Ideogram treats this phrase as the identity anchor. Fiction style pushes the model to lean into fantasy aesthetics while preserving the reference face.
Default speed · 3:4 · Auto style
the same woman in anime style, expressive eyes, soft cel-shaded coloring, outdoor cafe in spring with cherry blossoms drifting past, looking off-frame with a small smile
Anime variations are one of the most impressive demonstrations of identity transfer. The reference face translates into stylized form while preserving recognizable proportions, hair, and expression. Auto style works best for cross-genre jumps.
Default speed · 3:4 · Realistic style
the same woman as a 1940s film noir detective in a charcoal wool coat with the collar turned up, standing under a rain-soaked street lamp at night, smoky high-contrast black-and-white, harsh shadows across her face
Period and genre transformations work well — name the era, the mood, and the lighting. Realistic mode keeps photographic texture even when pushing into stylized noir compositions.
Default speed · 3:4 · Realistic style
the same woman in a tan utility spacesuit (helmet off, tucked under her arm), short auburn hair tousled, standing on a Martian ridge at golden hour with a small rover behind her, lens flare, photoreal sci-fi editorial
Sci-fi and concept-art variations land cleanly when you describe environment + costume + lighting separately. The character's signature features (hair color, freckles) survive even when wardrobe and scale change radically.
Spend the time getting one great reference: clean lighting, neutral backdrop, front or three-quarter view, sharp eyes, hair clearly visible, no occlusion. A clean reference dramatically improves consistency across every downstream variation.
Phrases like "the same woman", "the same man", "this character" signal to Ideogram that identity must be preserved. Start every variation prompt with that phrase, then describe the new scene/wardrobe/style.
Realistic for photoreal output (period photo, sci-fi editorial, fashion). Fiction for stylized/fantasy work (armor, magic, illustrative). Auto when you're not sure — it picks based on the prompt. Mismatched style modes produce hybrid outputs.
Turbo for rapid iteration (5+ variations to pick from), Default for production work (best identity-vs-quality tradeoff), Quality for hero shots and final renders. Most character-set work is Default-tier territory.
List the outfit explicitly ("wearing a charcoal wool coat with the collar turned up") rather than implying it. Ideogram preserves identity but happily restyles wardrobe — explicit wardrobe direction prevents random default outfits.
Magic Prompt expands your wording into richer descriptive prose before generation. Useful when prompts are sparse; turn it off when you have precise wording you don't want rewritten.
| Setting | Values | Notes |
|---|---|---|
| Reference image | Required (one image) | Cleaner reference = more consistent variations. Front or three-quarter view, sharp eyes. |
| Rendering speed | Turbo · Default · Quality | Trades speed for identity fidelity. Default is the production sweet spot. |
| Style type | Auto · Fiction · Realistic | Realistic for photoreal, Fiction for stylized/fantasy, Auto picks based on prompt. |
| Aspect ratio | 1:1 · 16:9 · 9:16 · 4:3 · 3:4 · 3:2 · 2:3 | Seven presets covering square through portrait/landscape extremes. |
| Resolution | Auto via aspect ratio · or fixed presets | "None" defers to aspect_ratio. Fixed pixel presets available for precise output sizing. |
| Magic prompt | Auto · On · Off | Optional prompt rewriting for richer detail. Off when your prompt is already precise. |
| Outputs | 1–4 per call | Each output counts as its own credit charge. |
LoRA training requires a dataset of 10-30 images, takes 20-60 minutes to train, and produces a model file you reuse forever. Ideogram Character needs one reference image and writes consistent variations on demand — no training, no model file. LoRAs win on extreme fidelity and aesthetic control; Ideogram wins on speed-to-character-set.