Tháp Korin Topic: Game, SLP

Summary

The video is a game-development discussion and live demonstration of an AI-assisted workflow called SLP. It covers multiplayer architecture, game-engine choices, contest ideas, and an early prototype for a subtle Vietnamese folk-horror game.

Game Ideas and Storyline

The speaker recommends making a game that feels distinctive rather than simply following familiar Halloween or gore-heavy horror trends. A Vietnamese setting and folklore could help a contest entry stand out.

The main concept demonstrated is a water-puppet-theatre story:

• The theatre has been neglected as people turn away from traditional folk art, so the puppets continue performing for an almost-empty audience.
• A living player enters and interacts with the puppets, who are surprised to encounter a real person.
• The audience’s shadowy figures are eventually revealed to be the spirits of people who came before.
• The player may answer reflective questions and take part in the performance—for example, drumming while the puppets respond to the rhythm.
• Choices, such as whether to return the next day, shape the ending and its sense of regret or nostalgia.

The intended mood is quiet, eerie, and nostalgic, without relying on jump scares, blood, or gore. The speaker suggests using atmosphere, old-fashioned imagery, sound, lighting, and environmental changes to build unease.

The prototype was still being developed during the stream. The speaker discussed refining the puppet scenes, audience, clouds, trees, moonlight, and animations. They also showed an airplane-shooting game and referred to a networked racing demo.

Gameplay and Multiplayer Development

The speaker argues that simple, standalone HTML games are less compelling than games where players can interact online.

For multiplayer games, the speaker favors a server-authoritative model: clients send inputs, while the server decides the game state. This is presented as safer than relying on a player-hosted server or a peer-to-peer setup.

Fast-paced games may need client-side prediction to keep controls responsive. The client predicts upcoming frames, then reconciles with the server; incorrect predictions may require rolling back to the authoritative state.

The speaker notes that prediction becomes more difficult in a 3D flying game than in a racing game constrained to a flat plane.

A multiplayer shooter concept is also described: players choose classes, split into teams, and use standard attacks, rockets, special attacks, and unlockable active skills.

The speaker claims their existing backend can support many rooms on a modest VPS and estimates that building a game backend could take around two weeks. These are the speaker’s claims, not independently verified benchmarks.

Development Strategies and Tips

• Choose a distinctive, achievable concept. A card game was suggested as a potentially manageable contest entry; a Vietnamese folk-horror theme was another option.
• Use assets thoughtfully. AI-generated art can help with concepts and components, but the speaker recommends guiding the style carefully and separating artwork into layers if it needs animation. More polished character art may require a proper modeling and animation workflow.
• Treat audio as part of the atmosphere. The speaker demonstrated AI-generated sound and suggested using sound to evoke the theatre’s setting without making it visually grotesque.
• Compare technical options through research. SLP can be asked to investigate relevant libraries—for example, physics engines when a game needs collision, gravity, friction, or rollback support.
• Test performance, especially in web games. The speaker highlights GPU use, texture filters, camera transforms, and image-processing work as areas worth checking.
• Choose a suitable browser-based 3D engine. The speaker favors PlayCanvas, describes Babylon.js as another option, and treats Three.js as a lower-level 3D engine rather than a full game engine. The spellings in the subtitles may be imperfect.
• Structure AI-agent workflows carefully. The speaker’s broader SLP workflow uses a lead agent, specialist agents, and sometimes a supervisor. They recommend giving agents clear roles, limiting unnecessary tools, and keeping project documentation short and current.
• Review work from multiple perspectives. The speaker describes using two reviewers—sometimes powered by different models—and comparing their areas of agreement and disagreement instead of accepting one review uncritically.
• Question mechanisms before optimizing them. The speaker’s “Demon Law” framework is described as a small index of principles to consult when relevant, rather than a large collection of instructions loaded all the time.

Gamers and Sources Mentioned

The speaker refers to Toko, Hoang Duc, and Tuan Anh in connection with game-making or assets. Game and development references include PUBG, Dota, Magic: The Gathering, League of Legends, Genshin Impact, Valorant, Turbo Cast, Overworld, and a Three Kingdoms/Warring States game.

Tools and platforms mentioned include PlayCanvas, Babylon.js, Three.js, Unity, GPT/OpenAI, Codex, and Sam 3. These are references made during the discussion; the subtitles do not establish that all were demonstrated in the video.