If you've spent time using a video generator, you know the frustration "missing the physics", to control a realistic scene, a prompt just isn't enough. Some of the common actions in life as simple as opening an umbrella, opening a lock at the door - are commonly difficult in AI Video generation. You ask for a perfectly poured latte art heart, but you get art jut appears out from nowhere suddenly.
The AI interprets the references not just as a style guide, but as a roadmap for the transformation itself.
A few people aware actually you can control these sequences in your Reference Images.
I made two samples to show you how to leverage the power of references to create consistent, and believable "before and after" sequences here:
☕ Example 1: The Perfect Pour – Mastering Latte Art
Reference shared here: https://www.vidu.com/home/reference?params=eyJpZCI6IjMwNDAwMTgxODk4MDcyNzkiLCJpc0FjdGl2aXR5IjpmYWxzZX0=
Prompt: "Camera gently seeing the action of pouring milk into the coffee from no milk, slowly forms into a latte art in the coffee, frame end at[@latte1]."
Tips: you can even control the shape of the latte art in your reference:
🎨 Example 2: From Sketch to Masterpiece – Visualizing Creative Work
Reference shared here: https://www.vidu.com/home/reference?params=eyJpZCI6IjMwNjExNzkyMDMxMDE4MzMiLCJpc0FjdGl2aXR5IjpmYWxzZX0=
Prompt: "a person's hand is drawing the sketch of santa [@santasketch1]starts from light then filled with color, then santa comes to live, smiles and turns looking at the side. static camera.."
Tips: For more detailed drawing recommend to use 8s just to draw.
You can even control the angle/actions of the Santa art in your reference:
💡 Pro Tips for Sequence References
Use 3 consistent Images: While you can use fewer, using a Start, a Mid-Point, and an End (3 images). I usually ask 1. nano banana, 2. Vidu.com reference to Image (its free now until year end), to do the images, but! for some tricky case, u might need to further adjust in photoshop or mask it at leonardo.ai to make the exact after/end result to create the image.
Make sure you give it enough time: Please the 8s to do these sequence actions. These simple actions like drawing, opening doors etc are actually takes time to look smooth, then use Extend to do the next move will comes out with better realistic results.
Why not just use image to video: Reference to video sequence is able to just keep that part of your character + the action consistent! You can feed your character + the coffee reference to control the scene x2.
Try it out! Next time you need a challenging "before and after," skip the long, complex prompt and give the generator just the right references and it can feel so much easier.
🤯 Free! 100 credits free trial!
I know hands on experience counts.
I got a code from them u can get 100 credit free to start trying whether your ideal sequences works on them:
Code:RAINBOW
Link: https://www.vidu.com/login?invite_code=RAINBOW&utm_source=cpp&utm_medium=cppcode&utm_campaign=q2
At vidu ref to video is 12credit for 5s, 20credits for 8s, so 100 is quite a good start with no commitment to try.