Getting Started with UNI-1
UNI-1 is accessible directly through the Luma AI platform at lumalabs.ai/dream-machine. No local installation required — the entire model runs in the cloud.
How to access UNI-1:
- 1Visit lumalabs.ai and create a free account
- 2Navigate to Dream Machine and select the UNI-1 model from the model picker
- 3Enter your first prompt in the text field and click Generate
- 4Your image will appear in the output panel within seconds
Free accounts receive a daily generation allowance. Paid plans unlock higher resolution outputs, priority queue access, and commercial usage rights.
Writing Effective Prompts for UNI-1
UNI-1's reasoning engine understands complex, natural-language instructions. Unlike other models that require terse keyword prompts, UNI-1 benefits from descriptive, detailed descriptions.
Recommended Prompt Structure
"cat in forest"
"A tabby cat sitting on a mossy log in an ancient forest at golden hour, soft dappled light filtering through oak leaves, painterly impressionist style, warm amber tones, peaceful and serene mood"
Useful Prompt Templates
- Portrait: "Professional headshot of [person description], studio lighting, [background], photorealistic, 4K detail"
- Product: "[Product] on [surface], [environment] background, commercial photography, clean and minimal"
- Fantasy: "[Character/Scene], epic fantasy art style, dramatic lighting, highly detailed, concept art for [game/film type]"
- Architecture: "[Building type] exterior, [architectural style], [time of day], [weather], architectural visualization render"
Image Generation with UNI-1
UNI-1 supports both text-to-image and image-to-image generation. The model outputs at up to 2K resolution with precise prompt adherence.
Text-to-Image
Describe your vision in natural language and UNI-1 renders it. Works with complex multi-element scenes that trip up other models.
Style Control
Choose from 76+ built-in art styles — from photorealism to anime, oil painting to pixel art. Mix styles with weighted modifiers.
Resolution
Output at standard (512×512), HD (1024×1024), or 2K (2048×2048). Higher resolutions use more credits but deliver professional-grade detail.
Multi-turn Image Editing
UNI-1's most powerful feature is conversational editing — the ability to refine your image through follow-up messages, just like chatting with a designer.
Example Editing Conversation
Key advantage: UNI-1 remembers the full conversation context, so each edit builds on the previous state without losing coherence or restarting from scratch.
Advanced UNI-1 Techniques
Complex Scene Composition
Use spatial modifiers to control element placement: "in the foreground", "behind", "to the left of", "partially obscured by". UNI-1 handles up to 8 distinct objects with accurate spatial relationships.
Multi-image Combination
Upload 2–4 reference images and describe how to combine them. UNI-1 can merge styles, characters, and environments from multiple inputs into a single coherent output.
76+ Art Style Application
Specify styles precisely: "in the style of Claude Monet", "ukiyo-e woodblock print", "brutalist architectural rendering", or "cyberpunk concept art". Combine two styles with weighted syntax: "80% photorealistic, 20% watercolor".
Best Practices & Tips
Be specific about lighting
Lighting dramatically affects output quality. Always specify: golden hour, soft studio, harsh midday, candlelight, etc.
Use negative prompts sparingly
UNI-1 understands exclusions naturally. Write "without shadows" rather than adding a separate negative prompt.
Iterate in small steps
For complex edits, make one change at a time in multi-turn mode. Large single-step changes may cause unintended drift.
Leverage aspect ratio
Specify orientation for better composition: "portrait orientation", "wide cinematic crop", "square format for social media".
Name the camera
Add "photographed with a 50mm lens", "aerial drone shot", or "macro close-up" to control perspective and depth of field.
Reference real art movements
UNI-1 has extensive art history knowledge. "Art Deco poster style" or "Bauhaus geometric design" yields more precise style application than generic descriptors.