My Oshi Canvas

"The Process of Turning Imagination into Reality"

The Problem: Uncertainty in Creative Outcomes
- Current AI image generation technologies offer powerful features. However, creators often face a daunting barrier: "Prompt Engineering." The process of repeatedly generating images until a desired result appears feels less like "creation" and more like "gambling" on luck. We focused on this inefficiency where technology overshadows creativity.
The Goal: An Environment Focused Solely on Imagination
- Creators should be architects of ideas, not slaves to technology. We wanted to build an environment where one can precisely complete a character envisioned in their mind without struggling with complex prompt syntax or endless trial and error. Our mission was to strip away technical complexity and fill that space with the creator's intuition.
Source of Inspiration: Tools Should Resemble the Creator’s Hand
- MyOshiCanvas began with the belief that "the finest tools should be as intuitive as a creator's hand." By utilizing a familiar chip-based selection UI, we capture the user's intent. AI then acts as an architect, analyzing and organizing that intent into a sophisticated structure.

"An Intelligent System for Realizing Imagination"

Intuitive Design: Chip-Based Character Architecture
- Instead of complex text prompts, users select "Design Chips" that embody their intent—such as gender, personality, and atmosphere. The AI designs the visual elements based on these selections, instantly realizing the character as imagined.
AI Design Report: Collaborative Intelligence through Analysis
- Beyond just showing a result, we provide a deep analysis report explaining why the AI chose specific design elements (body type, pose, hair, expression, outfit, accessories). This builds a bridge of trust between the creator and the AI. It allows creators, even those without professional backgrounds, to focus on creative judgment through logical AI suggestions.
AI Identity: The Completion of a Character
- Naming and giving a story to a character is one of the most difficult stages of creation. MyOshiCanvas's AI suggests the perfect name and background story that fits the character's appearance. This transcends a mere image, completing the character as a unique identity—the "Oshi" (favorite) we encounter in great works of art.

"Sophistication Through AI Agents and Dogfooding"

Full-Cycle AI Agent-Driven Development
- The project was built 100% utilizing AI agents. Initially, we conducted repeated experiments with Gemini models to identify the possibilities and limitations of AI-driven image generation, verifying our core logic.
System Architecture & Interface Implementation
- Once the validity of the generation logic was confirmed, we developed the server-side operations and User Interface (UI) to process user selections in real-time. We designed a stable communication pipeline to ensure that user input is never lost or corrupted, even with a large number of custom chips.
Continuous Dogfooding & Quality Assurance
- We underwent a rigorous dogfooding phase, using the product ourselves under various conditions. This process allowed us to identify bugs, refine the user flow, and ensure a high-quality, functionally sound product.
Professional Product Identity Branding
- To emphasize that this is a professional tool, we focused on building a strong identity through landing page production, design refinement, and system log implementation.

"Breaking the Frames That Limit Potential"

Preconceptions About AI Agent Capabilities
- We had to look beyond AI as a simple output tool and explore how much an AI agent could actually contribute to the architectural process. Breaking the frame of traditional workflows and adopting an experimental approach to deep AI interaction was our greatest challenge.
Development Barriers for Non-Programmers
- Building a production-ready product, rather than just scripts, was a massive barrier for a non-programmer. Despite various experiments via Google AI Studio, we needed the conviction that "Vibe Coding" could actually lead to a viable commercial product.
Infrastructure Scalability & Technical Bottlenecks
- Initially, lacking deep knowledge of serverless frameworks and queue systems, the environment struggled to handle even a single user. We faced the challenge of finding technical solutions that could remain stable even with over 1,000 concurrent users.

"Creative Will Leading to Technical Achievement"

Proving the AI Collaboration Paradigm
- We treated AI not as a machine following orders, but as a partner to share opinions with. By proactively asking the AI questions and sharing intent, we achieved a level of quality that would have been difficult to reach alone.
High-Efficiency Launch via Solo Development
- By combining foundational knowledge across various fields with the power of AI agents, we handled a workload that would typically require dozens of people. We proved that in this era, anyone with a clear goal can challenge themselves without fear.
Infrastructure Optimization for Large-Scale Environments
- We filled our knowledge gaps by learning alongside AI agents in real-time. Based on this, we completely overhauled the queue system and serverless infrastructure, building a stable environment capable of servicing over 1,000 users.

"New Insights Beyond Technical Boundaries"

Creative Freedom Through Structured Constraints
- We learned that precisely designed options (chips) actually stimulate a user's intuition more sharply than infinite freedom (prompts). We confirmed that creativity flourishes more efficiently within a well-defined framework.
Active Interaction with AI Agents
- We realized that AI becomes a true partner only when the creator proactively shares intent and engages in dialogue. "Questioning" and "Conversation" are the keys to overcoming technical limitations.
Proving the "Impossible" Through Execution
- We are now in an era where technical gaps can be supplemented in real-time with AI agents. We experienced firsthand that with the will to challenge ourselves, anyone can build production-level infrastructure and products.

"Expanding the Ecosystem Focused on the Essence of Creation"

Convergence with Various Creative Fields
- We plan to expand into all areas requiring character design, including games, webtoons, animation, and VTubing. This will allow creators to focus on their core essence—fun for games, story for webtoons, and communication for broadcasters.
Extension into Backgrounds and Event Scenes
- Beyond character generation, we plan to add features for designing backgrounds and illustrating "Event Scenes." This will actively support creators in expanding their world-building for games and webtoons.
Animation Architecture for Vivid Motion
- We are preparing animation features to implement various poses and movements. Bringing characters to life through dynamic motion will be a powerful tool, providing the highest level of immersion in game and animation production.

Built With

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.