🌟 How do I keep characters, objects, and locations consistent?

The biggest giveaway in AI video is drift: a character's face changing between scenes, a product looking different each time, or a location that won't stay the same. Atlabs solves this with three top-bar tools, Characters, Objects, and Locations, that lock an entity's appearance and reuse it across every scene. Define it once, and it stays consistent everywhere it appears.

Keep characters consistent

Use Characters to define each person or creature that recurs in your video, their look, and identity, so their face and design stay the same from scene to scene.

1. Click Characters in the top bar.
2. Add or select a character, and set its reference appearance.
3. Apply it across the scenes where that character appears.

💡 There’s no limit on custom characters. You can add as many characters as your story needs, either by uploading your own images or by creating them with AI, and use them together in the same video. Each character keeps the same look and their own voice across every scene and every episode.

Keep objects consistent

Use Objects to lock a recurring item, a product, a prop, a logo'd package, so it looks the same every time it appears. This is especially useful for product ads where the item must match the real thing in every shot.

1. Click Objects in the top bar.
2. Add or select the object and set its reference.
3. Apply it across the relevant scenes.

Keep locations consistent

Use Locations to fix a setting, a bakery, an office, a street, so the environment stays coherent across scenes set in the same place.

1. Click Locations in the top bar.
2. Add or select the location and set its reference.
3. Apply it across the relevant scenes.

💡 Tips:

• Set up characters, objects, and locations early, before generating motion, so every scene builds on the same references. For an uploaded character, select Keep Uploaded Look if the image already matches your video's style; use the same visual style across scenes.
• In scene prompts, tag the relevant character, object, or location with @ so it is applied to that scene.
• If a scene drifts, regenerate its storyboard image from the Image tab and try a different image model. Users have reported better character and location consistency with Nano Banana 2 or Flux 2 Pro instead of Automatic; GPT 2.5 and Seedream 5 Pro are also alternatives to try. Results can vary by scene.
• Once the storyboard image is right, generate motion from that image. Some motion models may not read character tags directly. A text/image-to-video model that only animates a start frame may not use character tags; create the scene image with the character references first, or use a reference-generation model and add the characters as references.
• For seamless positions between consecutive scenes, generate or choose the still images for all scenes first. When adding motion, set the current scene's image as its first frame and the next scene's image as its last frame (scene 1: image 1 to image 2, scene 2: image 2 to image 3). Alternatively, use the last frame of one scene as the first image of the next.
• Apply the references and check the storyboard image in each scene where the entity appears; editing a scene image does not automatically correct every other scene. You can regenerate only affected scenes rather than the whole video.
• Combine all three for story-driven content (like animated or music videos), where the same cast, props, and settings recur throughout.
• There’s no limit on how many characters, objects, or locations you can create, so give every recurring person, prop, and setting its own entry.

Updated on: 10/16/2026