Przejdź do treści
Image Generator Parameters / Character Reference (--cref)
Universal parameters

Character Reference (--cref)

A character reference is an input image used by an AI generator to reproduce the same subject across different environments, poses, and camera angles. By locking key visual traits such as facial structure, hairstyle, and signature clothing, creators can build coherent storyboards, graphic novels, and digital campaigns. This eliminates the unpredictability of text prompts, ensuring your protagonist looks consistent from one scene to the next.

Also known as: character reference, --cref, --oref, Omni Reference, character_reference_images

Side-by-side test results

Character Reference (--cref) - reference character (fictional)
reference character (fictional)
Character Reference (--cref) - new scene: library
new scene: library
Character Reference (--cref) - new scene: park
new scene: park

Model: Ideogram 3.0 Character (przez API fal.ai); postać wzorcowa z FLUX.1 [dev]. Settings: 1 reference image, seed 1411, TURBO mode. Date: 14.11.2026.
Prompt: the woman is reading a book in a quiet library, photo / the woman rides a bicycle through a park in autumn, photo

The character in this test is fictional: I generated her separately as a clean base portrait. Across both new scenes, Ideogram preserved her red hair, freckles, and yellow jacket. Her face in the library setting matches the reference closely, while the bicycle shot in the park frames her from a distance, meaning likeness relies primarily on her hair and clothing.

Syntax across generators

GeneratorParameterValuesDefault
Midjourney*Edit Model (V8.1/V8.2); legacy: --oref + --ow (V7), --cref + --cw (V6/Niji 6)W V8.X Edit Model zastąpił Omni Reference i Character Reference. Edit Model nie działa z --tileEdit Model: up to 4 reference images; --ow 1-1000--ow 100
FLUXimages (FLUX 3, up to 10) / input_image..input_image_8 (FLUX.2)Images referenced in prompt by orderFLUX 3: 1-10; FLUX.2: up to 8-
Ideogramcharacter_reference_images (Ideogram 3.0 Character) / @img_1 images (4.5)Extra charge for character reference. ideogram-3-character endpoints (generate, remix, inpaint)3.0: 1 image (up to 25 MB), optional mask; 4.5: according to FAQ 1-5 images-
GPT Imageinput images for editingDocumentation states directly: characters and brand elements may not be consistent across generations--
Seedream*image (reference images)up to 10 (5.0 pro/flash) / up to 14 (lite, 4.5, 4.0)-
Qwen-Imageimage1-10 (2.1) / 1-3 (3.0)-

* Midjourney: docs.midjourney.com blocks automated scrapers, so values are sourced from search engine indexes. Please verify settings on the official Midjourney site before production use.

When to use it

Common mistake

Uploading photos of real people without explicit consent violates image rights, and most platforms will actively block recognizable faces. Another frequent mistake is using a reference image with a busy background or harsh directional lighting, which the model often bleeds into the new environment. For best results, use a clear portrait set against a neutral, evenly lit background.

Ready-to-use example

Midjourney: girl reading in a library --oref <link> --ow 100
Ideogram (API): character_reference_images with 1 image
FLUX Kontext: character image in input_image + description of new scene

Frequently asked questions

How do you maintain character consistency across multiple scenes?

Generate a single high-quality portrait against a neutral studio backdrop with balanced lighting, then use that image URL or file as your reference anchor for every subsequent prompt.

Can you change a character's outfit while keeping their facial features?

Yes. You can lower the character reference weight parameter so the tool focuses on facial geometry rather than attire, while explicitly describing the new clothes in your text prompt.

Why does the generator keep copying the background from my reference image?

Generators frequently absorb environmental colors and lighting cues from reference files. Crop your reference image tightly around the face and shoulders, or replace the backdrop with a plain white or grey fill before uploading.

Related terms

🛒 Sklep Kinetyka CapCut Pro 460 zł/rok Lovable Pro 450 zł/rok Replit Core 119 zł/rok Zobacz wszystko →
× ‹ powiększenie ›