How Embertold Makes AI-Generated Scenes and Sounds Free After the First Time

Every time Embertold generates an image, sound effect, or voice line, it gets smarter. Here's how our caching system works, and why it matters to you.

The Problem

AI generation is powerful but not instant. Generating a scene image takes a few seconds, generating a sound effect takes a moment, and each generation costs credits. If every player had to wait for (and pay for) brand new content every time (see how credits work), the experience would be slow and expensive.

The Solution: Similarity Caching

Instead of exact matching (which would rarely hit), Embertold uses similarity-based caching for images and sound effects. When the AI generates content, we don't just save the result, we save an embedding of the meaning of what was requested in our vector database, alongside the asset itself.

When new content is needed, the system first checks: "has something similar already been generated?"

• Scene images — If a player enters "a torch-lit stone dungeon with mossy walls" and we've already generated "a dark dungeon corridor lit by flickering torches," that's similar enough. The cached image is served instantly.
• Sound effects — A "heavy wooden door creaking open" matches a "creaking old door opening." Same sound, served from cache.
• Voice lines — These use a different rule: if the exact same line of dialogue, with the exact same voice and delivery settings, has been generated before, it's served from cache. No similarity fuzz here, an exact match only, since a single wrong word in a voiced line is much more noticeable than in a background scene.

How Similar is "Similar Enough"?

We use a high similarity threshold: 95% for scene images, 94% for sound effects. This means the cached content must be very close to what was requested. A "dark dungeon" won't match a "sunny meadow." But a "torchlit dungeon corridor" will match a "flame-lit dungeon hallway."

What This Means for You

1. Speed — Cache hits load near-instantly. No waiting for generation.
2. Free — Cache hits don't cost credits. Ever.
3. Gets better over time — As more players play, the cache grows. Content that's commonly encountered becomes permanently free for everyone.

The Virtuous Cycle

Every player's gameplay enriches the cache. The first player to enter a tavern generates the image. The second player gets it from cache, for free, in milliseconds. The more people play, the richer and faster the experience becomes for everyone.

This is why we built Embertold with caching at its core, not as an optimization afterthought, but as a fundamental part of the experience.

Play more. Generate more. The world gets richer for everyone. To see what caching enhances, read about Immersion Mode.