Feature request: Improve future GPT models’ ability to create complete videos through code — HTML, Canvas, and WebGL
I would like future GPT models to become substantially better at creating polished, complete videos through code, with the goal of matching or exceeding the quality of impressive code-generated video demonstrations associated with Claude Opus.
The capability I am asking for is this: given a short creative brief, the model should take responsibility for the concept, storyboard, visual design, animation, transitions, audiovisual timing, and overall polish, and turn them into a coherent finished film with minimal shot-by-shot direction from the user.
A concrete reference, with publicly available source code
A concrete example of the result I would like future GPT models to achieve or surpass is an abstract algebra promotional film:
- Video title: 抽象代数宣传片, also titled 结构之美 · 抽象代数.
- Platform: Bilibili.
- GitHub project name: abstract-algebra-promo.
- GitHub author: AnctyEnly453.
- Main source file: 抽象代数宣传片.html.
I am listing the names here because my account currently cannot include links.
The reference is approximately 4 minutes and 23 seconds long, with an opening, a title sequence, ten chapters, and a finale. Its visual implementation uses JavaScript, Canvas 2D, and WebGL, with animated typography, bilingual subtitles, and scenes organized around a shared timeline. The project also includes a Python-generated soundtrack, embedded audio in the HTML player, and a separate video export workflow.
Several transitions illustrate the level of creative integration I am looking for:
- A snowflake breaks into particles that form the title.
- Twelve symmetry operations become positions on a clock in the following chapter.
- A finite-field point pattern transforms into a QR-code pattern.
- The finale returns to the opening snowflake motif.
These are useful examples because the visual transitions connect the subject matter, the composition, and the narrative across scenes.
What I would like future GPT models to improve
1. Planning a complete film
Develop a clear opening, progression, climax, and ending. Maintain pacing and visual continuity over several minutes, with each scene contributing to the whole.
2. Turning the subject into meaningful visuals
Choose visual representations that communicate the actual topic. For educational or scientific content, preserve conceptual accuracy while making abstract ideas visually engaging.
The model should be able to decide which ideas deserve a diagram, a geometric transformation, a particle system, a typographic sequence, or another visual treatment.
3. Designing expressive animation and transitions
Produce intentional motion with good timing, easing, camera movement, and visual hierarchy.
Connect scenes through shared shapes, objects, transformations, or ideas, so that transitions contribute to the storytelling.
4. Coordinating the visual and audio design
Keep typography, color, composition, effects, music, and sound effects consistent with the overall style.
Use a coherent timeline to coordinate scene changes, motion accents, musical phrases, and sound cues.
5. Taking more creative responsibility from a short brief
Given a brief such as “Create a visually striking promotional film about abstract algebra,” the model should be able to develop the concept and implementation, produce runnable and editable code, inspect the rendered result, and refine it with limited user intervention.
I would like the model to make strong creative decisions without requiring the user to specify every shot, animation parameter, transition, and visual effect.
Scope of the request
The reference film provides a concrete quality target. The same capability should generalize to science explainers, educational animations, product introductions, promotional films, and motion graphics.
Its subject, duration, and technology choices are examples. Future models should be able to choose an appropriate implementation for the task while preserving the user’s ability to run, inspect, and edit the result.
A useful way to evaluate progress
A useful evaluation would give different models the same creative brief and comparable time and revision budgets, then examine both the generated code and the rendered video.
The evaluation should consider:
- Visual quality and composition
- Clarity and accuracy of the content
- Pacing and narrative structure
- Animation quality and transitions
- Consistency across scenes
- Audiovisual coordination
- Reliability and editability of the implementation
- How much detailed user direction was required
My desired outcome is simple: I describe the film I want in a short brief, and GPT takes substantial responsibility for turning it into a coherent, compelling, finished audiovisual work through code.