How Claude Creates Animation with Code: A Review of User Experiments

According to descriptions of user experiments, Claude can create animation and computer graphics through programming rather than through a separate mode for directly generating video pixels. The model receives a description or drawing, writes JavaScript code, and the computer executes it to produce objects, effects, and a sequence of frames.

The available materials contain no primary quotations from the experiments. The reported results should be considered a retelling of user observations, not independently verified model capabilities.

This distinction determines the meaning of the demonstrations. The user sees a finished video or three-dimensional scene, but the underlying process may involve programmatic graphics: geometry, coordinates, layers, lighting, timers, and movement rules.

The outcome depends on exactly what Claude did and what the user or an external program provided.

How code-based animation differs from video generation

In the comparison under consideration, a video model synthesizes a sequence of images, trying to preserve the characters' appearance and motion consistency between frames. In code-based animation, the model describes the scene through instructions. A browser or other software environment executes those instructions, calculates the positions of objects, and draws the frames.

In this scenario, Claude acts as a computer graphics programmer. According to descriptions of the available demonstrations, it can analyze an uploaded image and work with code, but the ability to see a picture does not mean it can directly generate arbitrary video-file content in the same way as a video model.

What experiments do users describe

In one recounted user experiment, Claude received a brief instruction and about two hours of work. The result was an animated film approximately 30 seconds long, devoted to human life from birth to old age. According to the experiment's author, the model did not use image or video generators but wrote JavaScript code to render the frames and play the animation. The primary quotation is absent from the available source, so this result cannot be considered independently confirmed.

Another experiment began with a drawing of a boy with a puppy on paper. According to the user's observation, the character approached the viewer, grew larger, and moved beyond the frame. The shadow moved with it, which, within this description, indicates work not only with image scaling but also with separate visual layers.

In another demonstration described by a user, Claude added a hole to the drawing through which a green onion could be seen, as well as a glowing energy sphere. The user reported reflected glow on the paper and surrounding objects. Such a result may be related to programmatically defined masks, gradients, transparency, and lighting, but the exact implementation cannot be determined from the retelling.

There were also reports of creating a three-dimensional model of a Tesla car with code. The description of this demonstration itself shows that the model could form geometry and a scene in programmatic form. It does not prove that Claude contains a universal three-dimensional generator: the model's quality depends on the chosen format, libraries, level of detail, and requirements for the result.

A short animated film made from code and ready-made materials

In one user experiment, a short film was described as being made in a style associated with Japanese director Makoto Shinkai. According to the retelling, the process took about two and a half hours. The production used several ready-made background images and external music, while the characters' movement, effects, transitions, and a substantial portion of the animation were created with code.

This is an important caveat. The result cannot be described as a video created entirely from scratch by Claude if it contains third-party background images and a music track. It is more accurate to call it a programmatic assembly of a multimedia scene in which, according to the experiment's description, Claude generated a substantial portion of the movement logic and visual editing.

According to the experiment's author, Claude independently chose the composition of the frames, the pace of scene changes, and the visual details. Recurring symbols included a red line, two divided hemispheres, and a palm. This is a subjective assessment of one specific work, not an official characterization of the model and not proof that Claude consistently carries out artistic direction without human control.

How Claude Creates Animation

In the review author's account, the process looks like a sequence of connected actions:

1. Task analysis. Claude identifies the characters, background, scene duration, order of events, and required effects.
2. Choosing a representation. The model describes objects through points, lines, curves, polygons, images, text layers, or elements of a three-dimensional scene.
3. Writing the code. JavaScript defines coordinates, sizes, colors, transparency, layer order, and state-update functions.
4. Calculating movement. A timer or loop changes object parameters over time: position, scale, rotation angle, light intensity, and degree of transparency.
5. Execution. The code runs in an environment available to the computer, after which the browser or graphics library renders the scene.
6. Checking and correcting. If the scene does not run or looks different, Claude analyzes the error, changes the code, and repeats the cycle.

This approach offers an important advantage: an individual element can be changed by a rule. For example, a character's movement speed is set by a parameter, while the shadow can receive the same coordinates and scale with adjustments to its vertical position and transparency. In direct video generation, a similar adjustment often requires resynthesizing the frames and does not guarantee that the other details will remain exactly the same.

There is also a limitation. A code-based scene requires the model to describe the geometry and logic correctly. An error in the coordinate system, layer order, or time handling leads to a failure, an object's jump, or incorrect occlusion. The user receives editable source code but must verify that it runs and check the visual result.

If the code works, that still does not mean the scene will look correct in every environment: the result depends on the browser, graphics library, and available files.

What the aquarium-scene comparison showed

In a user comparison of two Claude versions, the aquarium animation was tested. According to the experiment's description, the more powerful version conveyed the movement of the fish, the overlapping plants, and the depth of the scene better. This observation applies to a specific test scenario and does not replace independent research using identical prompts, the same number of attempts, and predefined evaluation criteria.

The aquarium scene is useful because it combines several difficult tasks. The fish must move along different trajectories, the plants must cover part of the objects, and depth must be conveyed through scale, perspective, shadows, and layer order. Improvement in such an example may be related to more precise code planning, better spatial analysis, or the characteristics of a particular operating mode.

In the transcript, the test version was called “Fable 5 / Opus 5.6.” This designation cannot be considered the confirmed name of a public model. User observations about a hidden Claude Opus 5.5 update, a new Opus 5.6 version, or the Claude Next model also remain unconfirmed until Anthropic publishes an identifier, documentation, and access conditions.

What is confirmed and what remains rumor

The review materials describe Anthropic's official documentation as a source of information about Claude's work with images, code, tools, and artifacts. According to the retelling provided there, it also mentions interactive materials, small applications, and the export of some animation artifacts to video. This information supports a programmatic and interactive scenario, but by itself does not prove that a specific user demonstration was performed by a new or hidden model.

Verification of available models should rely on the interface of the specific product and Anthropic's official Models API for the relevant key. Reliable signs of a public launch include a published model identifier, a description of capabilities, access rules, and pricing. Reports of fresh memes, unusually strong 3D graphics, or a single successful test do not confirm that the result belongs to a new version.

But what exactly did the user experiments show? Above all, according to the review author's assessment, they showed Claude's ability to write code for computer graphics, correct it, and assemble animated scenes from programmatic elements. They do not prove that Claude directly generates images or video, nor do they confirm autonomous creation of complex cartoons without human involvement.

Limitations of the demonstrations

• One successful run does not show that the result is stable on other tasks.
• The original prompt, libraries, settings, and number of failed attempts are not always disclosed.
• Ready-made backgrounds and music change the amount of work actually performed by the model.
• The quality of the video depends on the computer, browser, graphics environment, and available files.
• Mentioning a version in a user transcript does not replace official confirmation from Anthropic.
• A demonstration of editable code does not mean that the model creates a professional production pipeline without verification.

For a developer, the practical conclusion is quite specific: Claude can be used as an assistant for scripting, JavaScript animation, SVG, Canvas, WebGL, and simple three-dimensional scenes. The user specifies the result and checks the execution, while the model translates the description into a structure of objects and programmatic rules. For a cinematic video with consistent character appearances and complex texture, direct video models remain a separate class of tools.

Frequently asked questions

Does Claude really create video?

Claude can write code that creates and plays animation and, in some of the described scenarios, produce an animation artifact for export. This differs from claiming that a language model directly synthesizes every pixel of the video sequence.

Can a drawing be brought to life through Claude?

The described user experiments show this scenario: the drawing is used as source material, while movement, scaling, shadows, and light are added programmatically. The result depends on the quality of the source image and the accuracy of the code, and the experiments themselves do not replace reproducible testing.

Does Claude create complete 3D models?

Claude can write code to describe simple three-dimensional geometry or a scene. The specific Tesla car demonstration confirms the possibility of such a programmatic experiment within the scope of a user observation, but it does not prove the existence of a complete universal 3D pipeline.

Has the Opus 5.6 model or Claude Next been confirmed?

No. In the provided materials, such versions appear as user designations and rumors. Without an official model identifier, documentation, and access conditions, they cannot be considered confirmed Anthropic products.

Who is Claude's code-based animation suitable for?

This approach suits developers, prototype designers, interactive-scene creators, and users who want to edit motion through code. For quickly creating a photorealistic video without programming, it is more convenient to consider specialized image and video generators.