Vela Mascot Studio

Generated characters drift. Ask a model for the same mascot twice and you get cousins: a different face, a different palette, a different age. That drift is the whole problem this studio exists to solve, because a character that changes between posts is not a brand character. Consistency comes from a model fine-tuned on the character itself rather than from longer and longer prompts, and from a character sheet that every scene prompt is written against. The result is one recognisable figure across a still image for the site, a talking-head video, and a short cinematic clip. Four agents run it on a single durable workflow. A director turns a topic into scenes. A prompter turns each scene into a prompt that names no vendor. An audio agent produces the narration and reports exactly how long it runs, so the visuals are cut to the voice rather than the other way round. An orchestrator assembles and publishes the result. Two rules keep it working. Only one file knows which vendor is in use, so changing the image or video model is a single change rather than a rewrite. And the pieces of a scene are joined by the scene's own id, never by position in a list — the alternative is one scene's narration playing over another scene's picture, a bug that is obvious on screen and invisible in the code.