Here are the models I used:
- DeepSeek V2 Chat: design ideas, prompts, descriptions and keywords.
- Flux.1-schnell: image generation.
- Stable Diffusion x4 upscaler: upscaling to 4096x4096.
- OpenAI text-embedding-3-small model + FAISS index: finding similar designs.<p>Total resource usage:
- 350 million DeepSeek V2 tokens (through OpenRouter).
- 1800 H100 GPU hours (through Nexius).
- Two weeks on my desktop workstation (generating PDFs and ZIPs, uploading to the server).
- 4.5 TB of disk space for everything.<p>Enjoy!