How I Created a Complete Anime Scene from 3D Model Motion 🎬⚔️🐉 This time

Prompt

How I Created a Complete Anime Scene from 3D Model Motion 🎬⚔️🐉

This time, I ran the experiment with GPT‑6 Astra.

I wanted to find out whether a basic 3D animation could be transformed into a cinematic anime scene while preserving the original motion.

The goal was not merely to create a video that looked similar. I wanted the anime warrior to perform the same actions at the same pace and with the same camera movement.

The final video places the anime version at the top and the motion source at the bottom, allowing viewers to compare them in real time.

I worked inside ChatGPT Work and combined several tools, each responsible for a different stage of the process.

Stage One: The Original 3D Animation

I started with an eight second video featuring two grey 3D models.

A sword fighter battles a large monster. He dodges its attack, strikes back and eventually brings the creature down.

This video served as the motion source for the scene.

It defined the body poses, sword movements, distance between the characters, pacing and camera movement.

This is an important part of the process.

When you want a specific movement, simply writing “the character fights” in a prompt is not enough. It is better to provide a video that shows the system exactly what the movement should look like.

Stage Two: Establishing the Visual Language

Before generating the anime, I defined how the film should look.

I created a reference image featuring a consistent samurai character:

Black hair tied back

A dark blue Japanese outfit

Black shoulder armor

A red sash

A katana

I also established the design of the monster and the location.

The scene takes place in a Japanese village at dusk, with cool blue lighting, warm lanterns, light mist and mountains in the background.

The reference image became the visual identity sheet for the film. I returned to it throughout the process to preserve the same warrior, clothing and atmosphere.

Stage Three: Turning the First Fight into Anime

I uploaded the original 3D animation to Genjutsu as the driving video.

Alongside it, I added the samurai and monster image as a visual reference.

The 3D animation determined what the characters would do.

The reference image determined how they should look.

This created the first part of the film, in which the anime warrior performs the same combat actions as the grey 3D model.

It is important to understand that motion transfer does not always produce a perfect match for every finger and joint. The system may still change small details, but it preserves the overall structure of the action and the camera movement.

Stage Four: Building a 3D Continuation

Next, I wanted to continue the story.

After completing the fight, the warrior walks through the village. A dragon suddenly appears. They fight, and the warrior eventually manages to climb onto the dragon and fly with it toward the horizon.

I built a new 3D scene in Higgsfield 3D Jutsu that included:

A white sword fighter

A white dragon with four legs and two wings

A Japanese village

Houses on both sides of the road

A camera following the action

I created an animation lasting approximately twenty seconds and rendered it as a white model video.

After rendering, it became clear that the motion was not good enough. The body movements were stiff, climbing onto the dragon did not feel natural and the wing movements looked mechanical.

When I transferred this footage into anime, the stiffness was transferred along with it.

That was an important lesson: an attractive visual design cannot repair a weak motion foundation.

I therefore rejected this version of the continuation and did not use it in the final film.

Stage Five: Rebuilding the Continuation with Seedance 2.5

Instead of starting another manual animation, I used the original 3D video as the foundation for the continuation.

I uploaded it to Seedance 2.5 in forward video extension mode and asked the model to begin exactly from the final frame, where the monster is lying on the ground and the warrior is standing behind it.

I created a 24 second continuation and divided the action into clear time windows:

From 0 to 5 seconds, the warrior recovers and walks toward the village

From 5 to 8 seconds, the dragon arrives and lands in front of him

From 8 to 15 seconds, the fight takes place

From 15 to 19 seconds, the warrior grabs the dragon, climbs onto it and sits down

From 19 to 24 seconds, the dragon pushes off the ground, spreads its wings and takes flight

In the prompt, I emphasized that the system should preserve the white and grey models while creating natural human movement, planted feet, visible body weight and continuous action without sudden jumps or the warrior magically appearing on the dragon.

The result was a new white motion reference that looked like a continuation of the original 3D animation.

This section is a white reference video generated with an AI video model. It is not a 3D file that can be opened and edited in Blender.

Stage Six: Checking the Motion

Before creating the anime version, I reviewed the new motion video.

I focused on four key questions:

Does the walking look natural?

Is there clear physical contact during the fight?

Does the warrior actually climb onto the dragon rather than suddenly appearing on it?

Does he remain seated during takeoff?

Only after these actions appeared clearly did I move on to generating the anime version.

Stage Seven: Transferring the Continuation into Anime

I uploaded the white motion video to Genjutsu as the driving video.

I added three visual references:

The final frame of the first anime fight

The samurai character reference

The dragon design reference

I asked the system to preserve the entire motion sequence, including the footsteps, sword movements, contact with the dragon, climbing action, seated position, wingbeats and takeoff.

The white character became the established samurai.

The white dragon became a blue grey dragon with pale horns and copper colored wings.

The white environment became a lantern lit Japanese village beneath an evening sky.

Stage Eight: Editing the Complete Film

I edited the final video in Higgsedit.

In the first section, I placed the anime footage at the top and the original 3D animation at the bottom.

In the second section, I placed the anime continuation at the top and the white motion reference generated with Seedance 2.5 at the bottom.

Both layers were set to 24 frames per second and began at exactly the same time.

I did not change the speed of the anime or shift it in relation to the motion source.

I connected the eight second opening fight to the 24 second continuation, creating a complete film lasting approximately 32 seconds.

Stage Nine: Sound Effects and Music

Finally, I added a sound layer featuring footsteps, sword movements, impacts, dragon breathing, wingbeats and the sound of takeoff.

I placed cinematic music underneath the effects at a lower volume so that it supported the action without overpowering it.

I then balanced all the audio layers and exported two versions:

A full screen anime film

A comparison video with the anime at the top and the motion source at the bottom

What Did I Learn from the Process?

A video like this is not created with a single click.

It requires a combination of character design, visual development, motion creation, continuity testing, style transfer, editing and sound design.

The 3D animation provided the motion foundation.

Seedance 2.5 created the continuation.

Genjutsu transformed the movement into anime.

Higgsedit combined all the elements into one complete film.

The most important lesson is that the quality of the anime depends first and foremost on the quality of the underlying motion.

If the motion source is stiff, the anime will also look stiff.

When the movement is constructed properly, AI can be used to replace the characters, style and location while preserving the same cinematic action.

The transfer is still not perfectly identical in every frame, but this workflow provides far more control than generating an anime video from a text prompt alone.