Character Studio: AI Character Design and Casting

From character briefs and casting to full outfit images and revisions

AI AgentCodex图像生成多模态TypeScriptReactCloudflare WorkersD1R2View Project

Casting Candidates per Batch

Consistency Test Images

Image References per Revision

Project Summary

A private workspace for AI short-drama character development. It connects character briefs, casting, outfit generation, consistency checks and image revisions through Codex. Real tasks have verified nine casting candidates, eight consistency test images and a revision using both an identity master and the original outfit image. Results return automatically to the original project, while selection and acceptance remain manual.

Background & Challenge

Character briefs, prompts and generated images were spread across conversations and files. Repeated manual handoffs made casting comparison, outfit review and revision tracking difficult. Feedback also needed to become part of subsequent generation tasks without losing earlier images or character identity.

My Role

I defined the product workflow and acceptance criteria, then used Codex to develop and iterate the application. I prioritized stable identity references, automatic result delivery and explicit human review. Real generation tasks guided improvements to outfit creation, review controls and revisions based on user feedback.

AI Integration

The frontend manages character briefs, candidate comparison, outfit feedback and review. Cloudflare Workers provides the service layer, D1 stores projects, tasks and versions, and R2 stores original images. A local executor connects to the signed-in Codex client and uses its built-in image generation tool. Each task freezes the identity master, references, prompt and feedback. Stable task IDs support recovery after disconnections, and completed images return to the correct character while earlier images and versions remain available.

Implementation Process

Define requirements: map character development and clarify identity masters, outfits, tasks and reviews

Build the workspace: manage projects, characters, assets and version history

Validate casting: generate three candidates in each of three directions, with blind selection and side-by-side comparison

Develop outfits: generate full-body outfit images and automatically return eight consistency test images

Iterate from feedback: carry image-specific notes into the next prompt using the identity master and original image

Verify delivery: check real outputs, cloud files and version records, and support character package exports

Key Deliverables

A deployed private character design workspace

Nine automatically generated casting candidates across three directions, with comparison tools

Full outfit image generation and history management

Eight consistency test images with automatic result delivery and manual review

Saved image feedback and revisions driven by user notes

Persistent tasks, disconnection recovery and original output verification

Restorable JSON character packages and ZIP asset delivery

Reflection

Real use exposed several workflow gaps: tasks still required a return to chat, outfit specifications had no matching generated image, review controls were hard to find, and feedback did not reach the next generation. Iterations addressed these gaps with visible outputs, accurate task states and reusable feedback as acceptance criteria. Usage time and revision counts still need to be measured before making claims about efficiency or cost savings.

Interested in this technology?

Whether you are a potential partner, recruiter, or technical evaluator — start a conversation anytime.

Get in Touch