Using LLMs for Entertainment: A Guide

#product #oxlo #ai

Large language models have moved well beyond chatbots and autocomplete. Developers are now using LLMs as runtime engines for interactive fiction, procedural NPC dialogue, and multi-modal entertainment pipelines. These workloads are defined by long system prompts, persistent context windows, and agentic tool use, requirements that push token-based pricing to its breaking point. Oxlo.ai offers a request-based inference platform that flips this model: one flat cost per API call regardless of how much lore, history, or visual context you pack into the prompt.

Interactive Fiction and Branching Narrative Engines

Text adventures and visual novels rely on maintaining a coherent world state across many turns. A single request might include a 10,000-word setting bible, prior player choices, and current inventory. Under token-based billing, this overhead is charged on every single turn. Oxlo.ai charges per request, so your marginal cost stays flat even as the narrative deepens.

For these workloads, Qwen 3 32B provides strong multilingual reasoning for localized games, while Llama 3.3 70B serves as a reliable general-purpose narrator. If you are building puzzle-heavy adventures, DeepSeek R1 671B MoE offers deep reasoning without the per-token penalty for long prompts.

The following Python snippet uses the OpenAI SDK against Oxlo.ai to run a stateful dungeon master. The only change from a standard OpenAI implementation is the base URL and API key.

import openai

client = openai.OpenAI(
    base_url="https://api.oxlo.ai/v1",
    api_key="YOUR_OXLO_API_KEY"
)

system_prompt = """
You are the game master for a noir detective text adventure.
Setting: 1947 Los Angeles. Rules: Track inventory, reputation, and clues.
Never break character. Respond in second person, present tense.
Current world state: ... [5000 words of lore and history] ...
"""

response = client.chat.completions.create(
    model="llama-3.3-70b",
    messages=[
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": "I knock on the door of the Sunset Motel, room 14."}
    ],
    stream=True
)

for chunk in response:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

Procedural NPCs with Structured Output

Modern games need NPCs that react to dynamic world conditions and return machine-readable data. Oxlo.ai supports JSON mode and function calling across its chat models, letting you enforce schemas for dialogue, emotional state, and quest triggers.

Because Oxlo.ai has no cold starts on popular models, these calls feel instantaneous in live gameplay. You can also lean on Kimi K2.6 for advanced agentic coding when the NPC needs to reason about complex tool use, or GLM 5 for long-horizon agentic tasks where the character must plan across extended time scales.

response = client.chat.completions.create(
    model="kimi-k2.6",
    messages=[
        {"role": "system", "content": "You are a merchant NPC in a sci-fi RPG. Respond in JSON."},
        {"role": "user", "content": "The player shows you a stolen cargo manifest. How do you react?"}
    ],
    response_format={"type": "json_object"},
    tools=[{
        "type": "function",
        "function": {
            "name": "update_reputation",
            "description": "Adjust faction standing",
            "parameters": {
                "type": "object",
                "properties": {
                    "faction": {"type": "string"},
                    "delta": {"type": "integer"}
                },
                "required": ["faction", "delta"]
            }
        }
    }]
)

npc_reaction = response.choices[0].message.content

Entertainment is not text-only. A complete pipeline might generate a scene description, render it as an image, synthesize voiceover, and transcribe player audio commands. Oxlo.ai unifies these modalities under one API and one pricing philosophy.

Vision: Kimi VL A3B or Gemma 3 27B can interpret player-uploaded screenshots or hand-drawn maps.
Image generation: Use Oxlo.ai Image Pro, Flux.1, or Stable Diffusion 3.5 via the /images/generations endpoint to produce concept art or dynamic story illustrations.
Audio: Convert narration to speech with Kokoro 82M through /audio/speech, or transcribe player voice commands with Whisper Large v3 via /audio/transcriptions.