AI-Powered Image Generation System
Role
Senior Product Designer
Duration
2024 – 2025
Company
Click
Challenge
When I joined the team, our company was spending a fortune on frequent photo shoots for marketing materials, UI elements, and in-app banners. Not only was this costly and time-consuming, but it also led to inconsistencies in style and brand representation. Off-the-shelf generators were not an option either: they consistently failed at Uzbek specifics — facial features, traditional clothing, local food — and offered no real control over pose or composition. We needed our own way to create high-quality, brand-consistent images with realistic Uzbek faces, running locally so the data never left our environment.
Objectives
- Cost Reduction: Eliminate the need for expensive photo sessions.
- Brand Consistency: Ensure all images adhere to brand guidelines regarding poses, emotions, compositions, and colors.
- Cultural Representation: Produce images with realistic and recognizable Uzbek facial features.
- Security Compliance: Develop a solution that runs locally to safeguard data.
- User Accessibility: Create an intuitive interface for team members with varying technical backgrounds.
Process
1. Research and Model Selection
- Explored various AI models and tools, ultimately selecting Flux 1.0 for its superior image quality.
- Chose ComfyUI as the interface to build our custom workflow.
2. Prototype Development
- Developed a test workflow using the SDXL RealiVis 5.0 model due to initial hardware limitations.
- Integrated nested ControlNets (depth, pose, and canny) for precise control over image generation.
- Implemented face detailing and swapping techniques using IPAdapter to maintain facial consistency.
3. Securing Resources
- Generated sample images and presented them to the executive team.
- Secured budget for a high-performance local workstation based on positive feedback.
4. Full Implementation
- Upgraded to the Flux 1.0 model with the new hardware.
- Added advanced features like Object Separation and Direct Color Control.
- Developed a custom Figma plugin so generated images land directly in the design file.
5. Enhancing User Experience
- Created a user-friendly interface within Figma for easy access.
- Set up secure protocols for remote use.
- Enabled sharing of presets and workflows for team consistency.
Impact
- Cost Savings: Reduced expenses by eliminating the need for regular photo shoots.
- Brand Consistency: Achieved a uniform look across all marketing and UI materials.
- Efficiency Boost: Accelerated the image creation process, reducing turnaround time from weeks to hours.
- Empowered Team: Designers and marketers generate custom images without leaving Figma — no ComfyUI knowledge required.
- Cultural Representation: Produced images with realistic Uzbek faces and cultural attributes — traditional clothing, local food — that off-the-shelf generators couldn't get right.
Case Highlight: Custom Figma Plugin Integration
Problem: Designers needed to generate and place AI images without juggling multiple tools.
Solution: Developed a private Figma plugin acting as a bridge between designers and the AI image generation server. The plugin features an intuitive UI where designers can specify poses, emotions, compositions, and colors. Images are generated and inserted directly into Figma projects.
Outcome:
- Increased Productivity: Designers saved time by staying within one tool.
- Consistency: Ensured all team members used the same parameters.
- Ease of Use: Reduced the learning curve for non-technical staff.
Conclusion
The system outlived the novelty phase: image production moved fully in-house, and the plugin made it usable by people who had never opened ComfyUI. The bigger lesson for me was organizational — the budget for a ~$7,000 GPU workstation was won by putting real generated samples in front of executives, not by pitching the technology. The samples did what arguments couldn't.