Sodi – mascot & 200+ illustrations

The challenge

The learning platform needed a recognizable character – and hundreds of illustrations of him in one style across dozens of lessons. The core problem with generated images is instability: the same prompt easily produces a “different” hero. The task: keep the character consistent at the scale of 200+ images.

Approach & prompt process

1. 1

   Canon first, volume second: a character sheet was created – the reference with 10 canonical poses of Sodi, the single source of truth about the character.

2. 2

   New illustrations are never generated from scratch but via image_edit: the model receives the reference plus an instruction of what exactly to change – so the character doesn’t drift from image to image.

3. 3

   The prompt protocol is written down: how to describe the scene, what must never change (proportions, colors, features), how to keep the background consistent. Every prompt is assembled from this template.

4. 4

   Review with rejection: every batch was checked against the reference sheet; images with drifting features or style went back for regeneration – only what was stylistically indistinguishable made the cut.

5. 5

   Illustrations are tied to the content: each lesson’s scene was chosen to match its topic – no “image for the image’s sake”.

Outcome

• 10 canonical poses in the reference sheet

• 200+ illustrations in one style across all platform lessons

• A repeatable image_edit protocol: new images don’t break the canon

• The mascot became the visual foundation of the sodigi·learn brand