fal.ai: Reference-to-Video generation with Kling O1

Step-by-step

1. Navigate to the fal playground for Kling Omni reference-to-video
2. Enter a descriptive text prompt in the input field
3. Upload style images to guide the general appearance and hand-drawn cartoon style
4. Add elements for specific characters or objects, providing a frontal image for each
5. Upload additional reference images for each element to provide different angles (e.g., back views)
6. Tag the images and elements in your prompt using '@image1' or '@element1' syntax
7. Specify camera movements and transitions in the prompt to utilize the provided reference angles
8. Click generate to produce the video

Options

• Reference specific images as start frames in the prompt
• Use multiple style images to blend aesthetic influences
• Tag multiple distinct elements to track different characters or objects simultaneously

Watch out for

• Style images guide general appearance only, not specific objects or characters
• Elements require both a frontal shot and complementary reference shots for the model to understand the subject accurately

Tips

• Use the reference images to provide views the camera will transition to, such as a back view for a 180-degree orbit
• Explicitly mention which image style to maintain for specific elements within the prompt to ensure consistency

Highlights

“even though the prompt was very descriptive and referencing seven different images, it's still done an amazing job”