Bot Avatar vs our bot looks
Lead: Dominik Scholz's Bot Avatar turns any description into a character, and it is cheap because a chat model writes only outline geometry. We have more looks and a better animation system, but we have no "describe it and get it" option for free. I recommend we add that option as a new source for our existing builder.
What the posts show
The author made this tool because Grok Bot does not let you create new characters. It has 7 starter objects, 11 colours and 3 moods (Idle, Thinking, Happy). You can drag a character to turn it, and the eyes follow the cursor. It runs on X as a player card.
How a character is made
The tool calls no model and has no AI cost. The model never picks colour, style or motion, so every result looks like one family.
The three ideas worth learning
| Idea | How it works | Why it matters || --- | --- | --- || Geometry only | Reply holds id, name, one SVG path, eye x, y, spacing and scale | Short reply, no markup to trust, easy to validate || Strict rules | Canvas is 100 x 100, with limits on safe area, path size, body size and eye room | A bad reply is rejected, not shown broken || Shape sets the 3D | Each clockwise subpath is one part. A symmetric part turns round, and one merged outline turns as a slab | The model controls the 3D with no 3D data |Measured on the live page: the Controller is 29 stacked SVG layers with CSS translateZ, more tightly spaced at the edges. The page has no canvas and no WebGL.
What we have today
| Capability | Bot Avatar | Allternit || --- | --- | --- || Any object from a description | ok | no || Works with no AI cost | ok | ok || Pick from ready characters | 7 | 24 pack folders || Build from parts | no | ok || Colour choice | 11 | 9 + custom || Material finishes | no | 6 || Accessories | no | 12 || From a photo | no | ok || Animation states | 3 moods | 9 sheet rows || Bot gets the look on all surfaces | no | ok || Drag to turn | ok | warn || Eyes follow cursor | ok | warn || Bot picks its own look from chat | no | no |Our drag-to-turn works only in the builder preview. I did not check how far gaze reaches in each pack. bot_create in worker_bot_tools.rs sets no avatar on a new bot.
Our Originals are the closest match
They have the same look: one flat colour and two pill eyes. They are fixed abstract shapes, so a user cannot ask for a mug. Bot Avatar's advantage is the object silhouette, not the style.
What I would build
- Do it in the app: the user types a description and gets a character, with no copy and paste.
- Keep the saved shape:
config.avatarholds the portrait, the sheet and a recipe with the path. Edit reopens it. - Keep a local path: a user with no model access still gets Build a character.
- Ship on Desktop, ai.allternit.com and the m.allternit.com PWA in one change, with docs.
Not tested: I did not run their prompt through a model, and I did not prototype outlines in scene.ts. Our docs list extrusions there, but a step that converts SVG paths to shapes may be new. The model cost per character is also not measured.
Licence limit
I found no public repo or licence for this tool, so treat its code, prompt text and path data as all rights reserved. The ideas are free to use. Our builder docs also state that no third-party avatar-maker was consulted, so this work needs its own docs wording.