AniSketch

Inspiration

AniSketch started from a real bottleneck in anime production: there are not enough experienced cleanup artists, while many junior artists submit rough drafts that still need heavy revision. We saw talented people with creative potential getting stuck in repetitive rework cycles, and senior artists losing time that should go into expressive decisions.

For Track 5, Creative Flourishing, this felt like a clear opportunity: use AI to protect and expand human creativity, not replace it. In our view, a healthy studio pipeline is one where humans remain the authors, and tools reduce friction.

What we built

We built AniSketch, a platform that automatically cleans up rough anime drawings into review-ready drafts.

The workflow is designed to stay human-led:

1. Junior artist uploads rough work.
2. AniSketch generates a cleaned draft.
3. Senior artist reviews, edits, and approves.
4. Final creative ownership stays with the artist team.

Technically, we built a web-based frontend and a Python backend for model inference and job orchestration. The system supports GPU execution and asynchronous job processing so artists can continue working while results are generated.

What we learned

We learned that good creative tools are less about perfect automation and more about preserving intent. A technically cleaner line is not always the artistically correct line. The most useful outputs are the ones that keep style, gesture, and character personality intact.

We also learned that product design is an ethical choice. We intentionally positioned AniSketch as a copilot for artists, not an autonomous creator.

Challenges

Our biggest challenges were:

• Balancing cleanup strength with stylistic fidelity.
• Avoiding over-smoothing that erases personal drawing voice.
• Making outputs predictable enough for production use.
• Managing heavy model workloads and queueing on limited compute.

Another challenge was defining success metrics beyond speed. We tracked both efficiency and human satisfaction:

$$ Creative\ Value = \alpha \cdot Time\ Saved + \beta \cdot Artist\ Satisfaction + \gamma \cdot Style\ Preservation $$

This helped us avoid optimizing only for throughput.

Ethical considerations

For the Track 5 ethics prompts, our approach is:

• Humans stay creators: AniSketch assists cleanup, while artistic decisions remain with people.
• Enhancement over shortcutting: we reduce repetitive labor but keep the core creative process human.
• Voice preservation: we tune for style consistency so outputs do not homogenize artistic expression.
• Creative labor respect: the tool is aimed at skill development and mentorship workflows, not replacing artists.

How we built it

We implemented a full-stack prototype with a FastAPI backend, web client interface, model pipelines for image editing and generation, and scheduler-based job execution for production-like workloads. We also integrated iterative development support with Claude Code to accelerate engineering, debugging, and refinement.

Built With

• claude
• css
• diffusers
• html
• javascript
• python
• pytorch
• qwen
• transformers