How to Make Viral Animated Shorts Using Claude (Full Workflow)

Summary

The video demonstrates a workflow for making animated shorts with Claude and code rather than an AI video generator. The target style combines flat 2D characters with a 3D environment, comic-like outlines, expressive lighting, and smooth camera movement.

Reference Analysis and Style Guide

• The creator had Claude examine 20 reference shorts—about 18 minutes of footage and more than 32,000 frames—at 30 frames per second.
• Claude transcribed the original English audio and identified a recurring script structure: an attention-grabbing paradox or unusual claim, escalating details, an explanatory pivot, and a final payoff or twist. The creator also observed a fast narration pace and frequent use of “You see?” to introduce explanations.
• The shorts’ visual traits included saturated colors, warm highlights and cool shadows, outlined 3D scenery, and camera moves that often travel between moments instead of cutting.
• To make the analysis measurable, Claude generated contact sheets and tools to inspect cuts, image shifts, brightness, contrast, and saturation. The creator cautions that the initial camera-tracking tool missed very slow movements: later analysis found motion in about 65% of frames, rather than the tool’s initial estimate that the camera was nearly still in 80%.

Building the Short

• The creator chose a story about soldiers marching in step and causing a bridge to collapse, after rejecting an idea that overlapped with a reference short. Claude checked the story’s factual details before writing the script.
• Early tests used Blender, but production shifted to Three.js for building the 3D scene with code. Remotion was used to render the animation frame by frame.
• The scene used free 3D assets and procedurally created textures. Claude built multiple character expressions, a bridge that bounced increasingly under the soldiers’ steps, and a custom edge-detection effect intended to add black ink outlines.
• The creator checked still frames before rendering and caught issues such as misplaced cameras, unsuitable colors, and objects obstructing characters.

Iteration, Diagnosis, and Agent Workflow

• The first version looked less polished than the references. The creator identified weak or missing outlines, flat colors and textures, insufficient background blur, and character proportions that did not match the target style.
• Claude organized the analysis into seven specialist agents covering color, outlines, lighting and shadows, textures, characters, depth and framing, and motion. Another agent handled sound, and four more worked on revisions.
• To prevent agents from overwriting one another’s work, each received a separate project copy and responsibility for specific files. Their changes were merged afterward.
• The agents found concrete problems: a blur step was wiping out the outline effect; camera movement was too abrupt compared with the slower reference glides; the shadows and dark tones were off; and the characters’ proportions differed from the examples. They also corrected some of the creator’s initial visual judgments.

Tutorial and Resource Information

• The creator says a free Claude “skill” containing the workflow, rules, and fixes is available in a GitHub repository, along with 5,000 reference images. The creator says it will be updated over time.
• The video’s main practical advice is to have Claude measure visual differences rather than rely only on subjective impressions. When coordinating multiple agents, the creator recommends dividing project files as well as tasks.
• At the end, the creator previews a separate planned tutorial about making realistic 3D videos, including face animation, character rigging, world-building, and reducing rendering time. Those techniques are previewed rather than demonstrated in this video.

Main Speaker and Source

Mr omi is the video’s creator and narrator. Claude and its analysis and agent workflow are discussed and shown through the creator’s account.