Welcome to DreamBees MLX Studio! This guide is designed for artists, designers, and non-technical creative users who want to create stunning AI artwork locally on their Mac without writing code or subscribing to cloud services.
Open DreamBees Studio from your macOS /Applications folder or run npm run dev.
On your first launch, the app automatically presents a touchless onboarding setup:
- ⚡ Metal GPU Detection: Automatically detects your Apple Silicon chip (M1/M2/M3/M4).
- 🐍 Silent MLX Engine Setup: Installs the rendering engine quietly in the background.
- 🧠 Model Verification: Pre-checks models like Z-Image Turbo (6B) and FLUX.2 Klein (4B).
- ✨ Zero Configuration Required!
- 🚀 Z-Image Turbo (6B): Sub-second synthesis, crisp prompt fidelity, 6B DiT architecture (Recommended for all Macs).
- ⚡ Sana 2.0 Sprint: High aesthetic clarity, studio color grading (~0.4s – 1.2s).
- 🎨 FLUX.2 Klein (4B / 9B): Photorealistic spatial depth, micro-texture optics.
Type a prompt intent like "A majestic puffin perched on an Atlantic cliff during golden hour", or press ⌘⇧V to dictate your prompt using Voice Dictation!
- Model Selector Dropdown: Switch between Z-Image Turbo, FLUX.2 Klein, Sana 2.0, Wan 2.1, Lumina 2.0, and Stable Diffusion 3.5.
- Intent Chat Input: Type plain-English prompts into the multiline chat box. Press Enter to send (or
Shift+Enterfor multi-line prompts). - Voice Dictation (⌘⇧V): Click the microphone button or press ⌘⇧V to speak your intent aloud in any language.
- Gemma 4 Model Card Prompt Matrix:
- Collapsible Reasoning Drawer: Click Show Reasoning to see Gemma 4's thought process notes.
- Model Card Rewritten Prompt: View the prompt formatted for your chosen model with a 1-click Copy Prompt button.
- Live Latent Stream & Telemetry: Watch neural previews and step telemetry (
it/s, step progress, VRAM usage) update in real-time as your artwork is synthesized. - Action Toolbar: Download high-res PNGs, copy intents, or expand artwork into full-resolution Lightbox view.
| Shortcut | Action |
|---|---|
⌘ + ⇧ + V (or Ctrl+Shift+V) |
Toggle Voice Dictation |
| Enter | Send Prompt & Generate Image |
| Shift + Enter | Insert New Line in Prompt Input |
| Esc | Close Lightbox or Selection Dropdown |
| Unified RAM | Recommended Model Cards | Generation Latency |
|---|---|---|
| 8 GB RAM | Z-Image Turbo (6B), Sana 2.0 Sprint | ~1.2s – 4.5s |
| 16 GB RAM | Z-Image Turbo (6B), FLUX.2 Klein (4B), Wan 2.1 | ~8.0s – 12.0s |
| 32 GB+ RAM | FLUX.2 Klein (9B), SD 3.5 Large, All Models | ~5.0s – 10.0s |
No! Once model weights are auto-provisioned, DreamBees MLX operates 100% offline. You can create artwork on an airplane or without Wi-Fi.
No! Rendering runs with low process priority (nice -n 10) and Ironclad VRAM memory capping (25%–35% system RAM max, strictly reserving 65%–75% RAM for macOS), ensuring your desktop, browser, and macOS animations remain 60 FPS fluid.
All images are saved locally at ~/Library/Application Support/DreamBees Lite/generations/. Click Clear or Download at any time to manage your files.