Google Whisk AI solves the character consistency problem by using a three-input visual blending system (Subject, Scene, and Style). Powered by Gemini and Imagen 3, Whisk allows you to "lock" a character’s face and features in the Subject slot, changing the environment and lighting across prompts without altering their identity.
Here is the step-by-step workflow to generate consistent AI characters using Google Whisk.
Phase 1: Generate a Clean Base Character
To prevent the AI from confusing your character's clothing or face with environmental elements, always build your base reference image on a clean, isolated background.
- Clear Input Slots: Open Whisk in Google Labs (
labs.google/fx) and ensure the Subject, Scene, and Style reference boxes are completely empty. - Prompt the Base Character: Enter a detailed prompt describing your character against a plain backdrop.
- Example:
Full-body shot of a 28-year-old female architect with short dark curly hair, green eyes, wearing a simple white t-shirt and jeans, plain solid grey studio backdrop, neutral lighting.
- Select and Save: Choose the cleanest generated output where the face and body proportions are sharp. Save this image to your device—it serves as your core Subject Reference.
Phase 2: Lock the Character and Change Environments
Once you have your base image, lock it into Whisk's multi-input framework to render the character across multiple scenes and artistic styles.
- Load Subject Reference: Drag and drop your base character image into the Subject input box on the left panel.
- Set the Scene: Define where your character is located using one of two methods:
- Text Scene Prompt: Type the environment into the main prompt bar (e.g., Walking through a crowded Tokyo neon street at night, wet pavement, cinematic atmospheric light).
- Image Scene Reference: Drag a photo of an environment into the Scene box.
- Apply Visual Style:
- Select a built-in Whisk style preset (e.g., Cinematic, Photorealistic, 3D Render, Comic Book), or upload a style reference image into the Style box.
- Fine-Tune Reference Controls: Enable the Precise Reference toggle on the Subject slot to strictly lock facial geometry and avoid random face drift across generations.
- Generate: Click Generate. Whisk extracts the character from your Subject image, applies the environment from the Scene, and renders the entire composition using the active Style.
Core Input System Overview
| Input Box | Function | Best Practice |
| Subject | Controls character face, hair, and physique | Use high-resolution images with plain studio backgrounds and clear lighting. |
| Scene | Defines background, environment, and context | Keep environment descriptions detailed regarding lighting and time of day. |
| Style | Specifies artistic medium and color palette | Use uniform style references across a storyboard to maintain aesthetic flow. |
Tips for Best Consistency Results
- Avoid Over-Prompting Subject Features: When writing prompts for new scenes, do not re-describe the character's facial features in the prompt text. Describing the face again can cause the model to conflict with the locked Subject reference image.
- Specify Action & Expression via Text: Use the main prompt bar purely for dynamic actions, clothing changes, or emotional expressions (e.g., Character sitting at a wooden coffee shop desk, laughing, wearing a black winter coat).
- Maintain Aspect Ratios: Select your output aspect ratio before generating batch images:
- 16:9 for film shots, storyboards, and widescreen video base frames.
- 9:16 for mobile screens, reels, and stories.
- 1:1 for avatars and standard social posts.
- Build a Character Reference Library: Save 3 to 4 angles of your character (front profile, 3/4 turn, full-body) on plain backdrops. Swap these into the Subject slot based on the angle required for specific scenes.
){kind=link}
0 Comments