Required parts and how to submit an original character

Besides the standard characters, you can also use your own original character design (Standard plan and above). From "Character" in the admin dashboard, just upload a ZIP of transparent PNG parts and you can preview and apply it. If you're commissioning an illustrator, you can hand them this page's specification as-is.

How it works

Rather than a single image, the character is shown by layering three transparent PNGs — "face (base)," "eyes," and "mouth." Motion is done purely by swapping images (there's no smooth morphing). While speaking, the mouth cycles quickly through the "a/i/u/e/o" mouth shapes plus the mouth matching the current expression; the eye/mouth combination changes with the mood of the AI's reply; and every few seconds the eyes briefly switch to the "closed" image to blink.

Most important point
Export every part at the same canvas size and the same face position. If parts are even slightly misaligned, the character will visibly jump around each time the image switches. The safest approach is to start from the "face base" and draw only the eyes/mouth on top of it (leaving everything else transparent).

Reference images

Sharing the reference images below with your illustrator can help convey the idea of the expression and mouth/eye variations (these are AI-generated references). For the actual delivery, please provide individual transparent PNGs split by part, not a single sheet like this.

Example of expression variations (reference image)
Example of expression variations (reference image)
Example of mouth/eye part variations (reference image)
Example of mouth/eye part variations (reference image)

Image specification

FormatPNG with a transparent background (alpha channel)
Size400×455px wide/tall (up to 800×910px at the same ratio). Every file must be the same size. 1MB max per file, 12MB max for the whole ZIP.
Number of files20 required files (listed below)
AlignmentExport every file on the same canvas, in the same position as the face base. For the eye/mouth parts, keep everything outside the part transparent and export at the same size as the base (don't crop the part down to its own bounds).
Stacking orderbase → eye → mouth
Eyebrows / blushInclude these in the "eye" parts (needed for the troubled, surprised, and shy expressions). Don't include eyes, mouth, eyebrows, or blush in the face base.

Required files (20 total)

Face base (1 file)

File nameContents
base.pngHair, outline, body, etc. — everything except the eyes, mouth, eyebrows, and blush. The eyes and mouth are layered on top of this.

Mouth — a/i/u/e/o (5 files)

While speaking, these mouth shapes are cycled quickly to create a talking motion.

File nameContents
mouth_a.png"a" mouth shape
mouth_i.png"i" mouth shape
mouth_u.png"u" mouth shape
mouth_e.png"e" mouth shape
mouth_o.png"o" mouth shape

Mouth — expressions (7 files)

File nameContents
mouth_neutral.pngNeutral — idle mouth when not speaking (slightly closed)
mouth_smile.pngSmiling
mouth_closed.pngClosed — fully closed mouth (used between words while speaking)
mouth_troubled.pngTroubled
mouth_surprised.pngSurprised
mouth_shy.pngShy
mouth_sad.pngSad

Eyes — expressions (7 files)

File nameContents
eye_neutral.pngNeutral (default state)
eye_smile.pngSmiling
eye_closed.pngClosed (used for blinking)
eye_troubled.pngTroubled (includes eyebrows)
eye_surprised.pngSurprised (includes eyebrows)
eye_shy.pngShy (includes blush)
eye_sad.pngSad (includes eyebrows)

The AI picks one of 6 moods for each reply — neutral, smile, troubled, surprised, shy, and sad — each with its matching eye and mouth switching together. ("Closed" isn't used as a mood; it's only for blinking and for the mouth between words.)

How to submit it (admin dashboard)

  1. Under "Original character" in "Character settings," download the template ZIP (the standard character exported with these file names — use it as a reference for your own replacement).
  2. Replace the parts with your own character's artwork, keeping the same file names and sizes, and pack them into a ZIP (it's fine to compress them inside a folder).
  3. Upload the ZIP (you'll need to confirm you hold the rights to the images). Any missing files, mismatched sizes, or files without transparency are flagged as errors immediately.
  4. In the preview, click the "neutral / smile / closed / troubled / surprised / shy / sad / a / i / u / e / o" buttons to check how it looks.
  5. If everything looks right, click "Use this character" to apply it to the widget.
This feature is available on the Standard plan and above. You must own (or have a license for) the copyright and any other rights to the images you upload. If you downgrade to the Light plan, the widget reverts to a standard character (your uploaded data is kept).
← Back to the guide