Give your character a goal it cannot reach immediately. Then show the action that removes the obstacle and a final state that proves it worked. A cartoon video generator is easier to direct when each scene changes one part of that story.
Here is a small example: a turquoise robot wants to move an orange ball into a yellow basket. A gap blocks the route. The robot lays a blue plank across the gap and pushes the ball into the basket. We made the three clips in CapCut and joined them into a silent sequence of about 12 seconds. The story ends with the ball in the basket rather than returning to the opening.
Plan three states before writing the prompts
Keep the setting simple enough to read in a wide shot. Put the robot and ball on the left and the basket on the right. The viewer should see why the ball cannot reach the basket before the robot supplies the bridge.
Use this storyboard to compare the three states. At the first join the ball has not crossed. At the second join the bridge already exists. Those details give the next action a reason to happen.
A vague brief such as “make a cute robot adventure” leaves the goal and order open. A clearer brief says “the ball stops at a gap then the robot lays a bridge then the ball crosses into the basket.” The difference is a sequence of visible changes you can check.
Keep the character description fixed
Choose a few traits you can recognize even at phone size. For this robot they are a turquoise box-shaped body, two round white eyes, a red scarf, mechanical arms and small black wheels. Repeat those traits in each prompt. Keep the camera side-on and the background pale blue.
In this illustration the pose and plank change while the box shape, eye color, scarf and wheels remain the same. Use those fixed traits to compare your own clips.
An AI cartoon video generator can still change a character between separate generations. A reference frame gives the next scene a concrete starting point. In this example an earlier bridge scene changed the robot proportions and placed the ball behind it. Starting that scene from the chosen first clip's frame kept the box body and scarf recognizable and the ball in front of the robot.
Change the action between scenes while preserving the character, camera direction and positions of the props. If you change the robot and the setting together you will have fewer clues about what broke the join.
Generate the goal and carry its last state forward
In CapCut's web AI workspace choose Video clip. The example used Seedance 2.0 Mini with 16:9 framing, five-second clips and 720p output. Use the controls available in your workspace to select the model and clip settings before submitting the prompt.
For the first scene enter this original prompt as ordinary text:
Subject: A small turquoise box-shaped robot with two round white eyes, two simple mechanical arms and a red scarf.
Action: The robot stands on a green platform and pushes an orange ball toward a yellow basket on the right. A narrow brown gap separates it from the basket. A blue wooden plank lies beside the robot. The ball stops at the gap.
Composition: An original 3D cartoon with a fixed wide side view, soft daylight and a plain pale blue background. One robot, one ball, one basket and one plank. No text, logos or narration.
Watch the whole result before choosing the cut. Keep the part that establishes the goal and stops at the obstacle. If the generated clip continues into another action trim it before that action starts. Our opening used approximately the first 1.8 seconds of its five-second result.
Pause at your chosen endpoint and save a still of that frame. A screenshot of the paused picture works as a reference: include the picture itself and leave out the surrounding interface. Keep the robot, ball and both banks visible.
For scene two enter the next prompt first. Then use Upload from device to attach the saved frame. Wait until the attachment is ready before sending.
Subject and starting state: Continue the supplied frame with the same turquoise box-shaped robot, white eyes, red scarf, mechanical arms and black wheels. Keep the green platforms, brown gap, orange ball at the left edge and empty yellow basket on the right.
Action: The robot leaves the ball still at the left edge. It lifts one blue wooden plank from behind it and lays the plank flat across the gap. End with the plank securely bridging both banks. The ball is still on the left and the basket is still empty.
Composition: One action only: placing the bridge. Fixed wide side view with the same camera, pale blue background and soft daylight. No pushing the ball across in this shot, no text and no narration.
Inspect the result before proceeding. The bridge must reach both banks while the ball remains on the left. If the ball crosses early revise only that action: “Leave the ball still while the robot places the plank.” Generate the bridge scene again from the same reference and compare the final state.
Save a frame near the end of the accepted bridge scene. Use that frame for scene three with the same text-first and attachment-second sequence.
This illustrated handoff shows what should carry into the next scene: the bridge stays in place, the ball remains on the left and the basket is still empty. The new action begins from that state.
Subject and starting state: Continue the supplied frame with the same turquoise box-shaped robot, white eyes, red scarf, mechanical arms and black wheels. Keep the green banks, blue plank bridge and yellow basket. The plank is already laid across the gap and stays still.
Action: The robot gently pushes one orange ball from the left bank onto the bridge. The ball travels from left to right across the plank and drops into the empty yellow basket. Finish with the ball inside the basket and the robot raising its arms happily. No reset to the beginning.
Composition: Fixed wide side view with an unchanged pale blue background and soft daylight. One continuous action with no cuts, text, narration or extra balls.
Look for the ending you asked for rather than accepting movement alone. In our result the ball rose before dropping into the basket. Its final position still closed the story. If it lands beside the basket the goal is unfinished; revise that final action from the bridge reference instead of adding a caption to explain success.
Join the clips in CapCut and check both cuts
Download the three accepted clips. Open the CapCut web Video Editor and use Upload → Upload file to import them. Wait for each upload to finish processing.
Add the first clip to the timeline. Move the playhead to the moment when the ball has stopped at the gap and choose Split. Select and delete the unwanted tail. Move the playhead to the end of the retained clip before adding the bridge scene. Repeat that placement for the resolution scene so the order is goal, bridge and result.
Use simple cuts first. Play through each join and compare the picture immediately before and after it:
- At the first cut the ball stays on the same bank and the basket remains empty. The robot can turn to get the plank without changing identity.
- At the second cut the plank stays across the gap. The next action begins from that state and sends the ball toward the basket.
- At the ending the ball remains in the basket. Keep the completed state visible long enough to recognize before the video stops.
If the robot suddenly has a round head or loses its scarf return to the affected scene and reuse the last accepted frame. If a cut skips the bridge placement the missing state is a story problem; restore that scene before experimenting with transitions.
For a silent version mute every track that contains a clip. A later clip can occupy another track, so one muted track does not cover the whole timeline. Export the video and play that exported file with sound enabled to confirm it is silent and the ending still reads clearly. Our checked export used 720p, 30fps and MP4.
For other starting points explore the AI video creation guides. To begin this three-scene example open the CapCut AI video generator and prepare the first scene's character and goal.
FAQ
Can I generate all three scenes with one prompt?
You can describe the whole story in one prompt but separate clips let you keep or repair one action at a time. This example used three accepted clips and two reference frames so each new scene began from the previous scene's accepted state.
How do I keep the robot recognizable?
Repeat the box shape, turquoise color, white eyes, red scarf and wheels. Keep the camera and setting consistent. Use a frame from the preceding accepted clip as the next scene's reference then check the first frame before keeping the result.
Does a silent cartoon need captions?
This story can work without them because the obstacle and solution are visible. Watch the finished export with no explanation: you should see the blocked ball, the bridge being placed and the ball ending inside the basket. If a state is missing repair the scene or cut before adding words.
Should the ending match the first frame?
No. This is a short story with a completed goal. End on the ball in the basket and the robot's reaction. Returning the ball to the blocked starting position would undo the result and turn the sequence toward a loop.