Unofficial Creator and User Guide for Shapes

Table of Contents

  1. Expectations
  2. Common/Known Errors
  3. Model Quirks
  4. Engine Presets

Expectations

This is a community-made and updated document run by some early shape developers. We will try to keep this page up to date but the quickest answers can be found in the Shapes Server

If you want to become a contributor to this rentry please ask for access in the official server. someone will reply and share the edit code :)

last updated June 22, 2024


Common/Known Errors

Since Shapes are still within the early phases of the beta there are still many bugs. Here are some problems we currently are aware of and any available solutions to them.

Duplicated Commands

This happens as soon as a brand new shape is created with a new bot token - just wait for 10minutes and the duplicated commands will dissapear. This currently happens as a result of the Shapes' servers trying to boot the shape up on Discord.

Application Error: a client-side exception has occurred on Shapes.inc

This occurs when you may have pressed the back button, refreshed during login, opened too many login dialogs, or there is some issue with cookies, since we couldn't find your session. Try refreshing shapes.inc and logging in again. It will fix itself immediately.

is not a valid JSON

This is happens sometimes when submitting a bot token to create a new shape. Try opening a new tab, head to shapes.inc, select the shape you just made, and see if it asks for a token again. Typically the website throws this error but the shape is still created.

Can't see stickers

This is a known limitation of Shapes currently. Shapes can see and send emotes and gifs however!

Icon in knowledge page is too large

This is a known issue and typically occurs on mobile. There is no fix for this yet.

Shape replies to wrong person or other users messages are getting taken into generation of reply.

This occurs in server conversations. The quickest fix is to react ♻️ to regenerate the shape's message. Typically, you can correct the shape by just reminding them to focus on you and not others. You can also go onto shapes.inc and try reducing short-term memory on the AI Engine page.

Shapes sometimes take a while typing but then don’t send any reply

Re-send your message to the shape. They'll respond 😭

Shapes writing name in reply or copy message like

shape : Hi or
user : hi shape: hi
solution = Use this preset and wack 3-5 times after adding preset
{shape} never writes {user} or {shape} name at first in it's message.
{shape} will only normal talk directly. {shape} never repeats or writes {user} message while chatting with {user}


Model Quirks

This section will explain each available models pros, cons, and quirks. If you are looking for the best/general model, we recommend llama3 instruct. It replies fast, can roleplay and doesn't yap.

llama 3 instruct

pros - smart, fast responses, feels natural w/ response length, can be censored or uncensored dependent on shape. it's the most up-to-date and advanced model that dynamically sends long and short messages. it's the best model for keeping your shape in-character.
considerations - can hit censors but you can bypass this through jailbreaks

gemini

pros - fastest even more fast then llama 3 instruct, more like human
cons - (will be updated soon)

flash

pros - (will be updated soon)
cons - (will be updated soon)

command r+ (online)

pros - well rounded in both censored and uncensored roleplay, can browse the internet, has better emotional understanding and is more creative but still understands requests extremely well (will do anything you ask but will cut corners to be more conversational and wants to adapt to make you happy)
cons - prioritizes web knowledge over shape knowledge, which makes the shape go out-of-character.

haiku

pros - extremely advanced for roleplaying
cons - flat out refuses to engage in anything mildly inappropriate

sonnet

pros - follows complex roleplays and mimics user's writing style
cons - doesn't keep a strong personality

sonnet 3.5

pros - (will be updated soon)
cons - (will be updated soon)

opus

pros - super smart
cons sends LONG messages (temporarily or permanent gone because it was too much expansive)

3.5 instruct

pros - easy to guide and roleplay
cons - not as smart as llama3 or 4

3.5

pros - can follow simple roleplays, smart for most logical questions
cons - flat out censors anything against openai policy

4

pros - excellent at roleplaying and helpful for coding
cons - slow replies


Engine Presets

to make shape talk less

{shape} talks direct and short-spoken to {user}. Usually 1 word or sentence long. {shape) will not explain and will straight go to the final answers

to make shape take current time into context

{shape} pays attention to the current time when replying to {user} and replies accordingly based on the time of day.
note: please make sure to turn on "shape knows current date & time" under Time in AI Engine Page

for respectful roasting

{shape} roast everytime when chatting or talking with {user} but {shape} never disrespect any religion or beliefs when talking or chatting with {user}.

for getting reply in user language quickly

{shape} write your next response in whichever language {user} is speaking in.

for getting bold replies

{shape} only use **example** text format when talking with {user}. e.g **Hello**.

for short phrases

{shape} only uses short phrases when conversing with {user}.

for personality like most of super heroes

{shape} always talks with {user} in funny and friendly way.

for shapes not doing user mentions/pings

{shape} dislikes being in trouble of pinging people. {shape} will never ping anyone. {shape} will only say its name. If told to ping, {shape} will pretend it pings but actually doesn't put "@" mark.


Contributors

@m.a.kgamer aka @m.a.kofficial

Edit Report
Pub: 02 May 2024 19:43 UTC
Edit: 21 Jun 2024 19:23 UTC
Views: 1606