| name | loci-encoder |
| description | Turn each target into a vivid, memorable image and place it at a station. Use when the user says "encode these items", "make images for my palace", "help me remember names and faces", "turn this list into pictures", or after memory-palace-architect lays out the rooms. Applies bizarreness, multisensory, exaggeration, motion, and dual-coding; fills the loci[] of palace.json. |
loci-encoder
The encoding step — the irreducible core of the method. A locus only works if its image is
impossible to ignore.
When to use this
The palace skeleton exists (from memory-palace-architect) and each station needs the target
converted into an image that sticks. For digits/dates/cards, route to number-memory first to get
an image, then place it here.
Instructions
-
One image per locus. For each target, craft a single image and bind it to the station's
physical feature — the image must interact with the feature (sitting on it, smashing it,
growing from it), not just sit nearby.
-
Apply the levers (use 2–3 per image, not all at once):
- Bizarreness / distinctiveness — make it strange; ordinary images vanish.
- Multisensory — add sound, smell, texture, temperature, not just sight.
- Exaggeration — oversize, multiply, or distort scale.
- Motion / interaction — dynamic images beat static tableaux.
- Emotion — funny, gross, or startling outlasts neutral.
- Dual-coding — keep the target word and its picture linked, so two routes lead back.
-
Encode meaning, not spelling. For a concept, picture what it does or means. For a word in
another language, use a sound-alike substitute image (keyword method) plus the meaning.
-
Names and faces — see references/names-faces.md: convert the name to an image, anchor it to
a distinctive facial feature, and place the pair at a station.
-
Keep the route honest. Place images in route order; don't reorder for convenience. Spacing
from the architect step must survive — if two images bleed together, make them more distinct.
-
Write back. Fill each loci[].image in palace.json: text (the image in words),
techniques (which levers you used), and optionally a prompt for palace-visualizer to render.
Refusals
- No bland, literal images ("a photo of the planet") — those are not encodings.
- Don't place two targets at one station to finish faster; that is where recall fails.
- Don't fabricate facts to make an image neater; encode the real target.
References
references/names-faces.md — the names-and-faces recipe.
spec/MEMORY-PALACE-METHOD.md §3 (the encoding principles and their sources).
Built on SIP · mind-palace-agent-skills · Memory Palace Method v0.1