I use the stable difussion sd_xl_base_1.0_0.9vae.safetensors models, throuth python fast api provide interface, GPU resources is 4060 display card with 8G display memory, 64GB memory with I5-14600KF pc, my web deploy on vercel, the problem switch english language display chinese , I will check this issue, thank you your feedbac.
What it is – Michelangelo.best is a free, no-sign-up Stable-Diffusion playground. You can paste any prompt or choose from 300+ curated templates (all CC-0) and get a 512×512 result in ~10 s (: RTX 3090, SD 1.5).
Why I built it – I wanted a friction-less sandbox for friends who love generative art but struggle with GPU install or prompt writing. The model is far from perfect, so opening it up for feedback makes more sense than hiding behind a paywall.
Tech stack – FastAPI + Stable-Diffusion-WebUI on the backend, simple Vue SPA in front. Generation requests hit /generate (REST) with an API key system I’m still refining. Front-end is deployed via Cloudflare Pages; images stream back via SSE for a “typing” feel. Source snippets are in the footer if you’re curious.
Road-map
Negative-prompt auto-suggester (LLM-based)
Batch generation & upscaling
Share-by-link with seed + params embedded
Ask – I’d love brutal feedback on UX, prompt quality, and any failure cases you hit. If you have ideas on reducing cold-start latency or a favourite SD sampler I should expose, let me know. Thanks!
• UI footprint
- Michelangelo: single lightweight page, loads fast on mobile
- Leonardo : rich canvas, layers, real-time editing
• Primary use case
- Michelangelo: quick experimentation / teaching / idea sketching
- Leonardo : production assets, team workflow, hi-res output
TL;DR – Michelangelo is a free sketchbook for testing prompts; Leonardo is the full studio for polished, commercial assets. They complement each other.
leonardo.ai is my goal, i will continue optimize my web and catch to .leonardo.ai
Btw when you switch to english language, there still are some text in chinese.
Bye, Mario.