Build a consistent AI character with reference images

The single biggest difference between a one-off AI image and a believable AI persona is consistency: results that are more likely to feel recognizable across a set. SomaLabs helps you get there with reference images. You assemble a small set of pictures that define your character, attach the character to a generation, and give the model visual references to follow across images, plus video on eligible plans - no fine-tuning run to set up or babysit.

Why references beat one-shot prompting

Describing a face in words alone leaves too much to chance - hair length can drift, jawline cues can soften, and important details can get lost. Reference images give the model concrete visual anchors, so your character is more likely to read consistently across a set instead of feeling like a new variant each time.

Choose your reference images

Add as many quality references as your plan allows, up to your per-character limit. More important than the exact count is the spread: give the model enough angles and lighting to generalize from, without feeding it noise.

• Vary the angle: front, three-quarter, and profile shots so the model has more identity signal when it tries new viewpoints.
• Vary the lighting: soft daylight, indoor, and harder light teach the model the face, not one specific exposure.
• Keep the subject clear: the character should be the obvious focus, in focus, and unobstructed.
• Stay consistent on the identity itself: core identity cues, same defining features - this is the signal you want to reinforce.
• Skip heavy filters and busy backgrounds: they bleed into results and dilute the identity signal.

Create with your character attached

1. Open the Designer or your Characters list and create a new character.
2. Upload your reference set and let the character save.
3. In Studio, attach the character to your prompt, then describe the scene, wardrobe, and mood you want.
4. Generate. The references help steer identity while your prompt steers the pose, setting, and outfit.

On video-enabled plans, you can use the same character and reference workflow when moving from portraits into image-to-video. The shared credit pool keeps that workflow in one studio, though video results can still vary.

Start building your character

A consistent character is the foundation everything else builds on. Create one on a free account and see how well the identity carries across your first few generations.