Home / Models / Uni-1 on Floyo
AI IMAGE GENERATION + EDITING
Run Uni-1 on Floyo
Reasoning-first generation. Create and Modify modes. Up to 9 reference images. 76+ art styles. Accurate text rendering in English and Chinese.
Run Luma AI's Uni-1 through ComfyUI workflows in your browser. No API key, no installs, no local GPU.
Resolution
Up to 4K
Reference Images
Up to 9
Art Styles
76+
Reasoning
Built-in thinking
No installation. Runs in browser. Updated April 2026.
What is Uni-1?
Uni-1 is the first unified understanding and generation model from Luma AI, released on March 5, 2026. It is a decoder-only autoregressive transformer that processes text and images in a single interleaved token sequence. Instead of matching prompts to pixels like diffusion models, Uni-1 decomposes your instructions, resolves spatial constraints, and plans composition before rendering.
Everything in Uni-1 starts with one question: are you creating something new, or changing something that already exists? Create mode produces new compositions inspired by references. Modify mode edits a specific input image while preserving its structure. This separation gives you clear control over whether the model treats your input as inspiration or as a source to preserve.
The reference system is where Uni-1 gets precise. You can provide up to 9 reference images, each with a defined role: style, character, composition, color palette, lighting, texture, or mood. You label each reference and the model treats it as having authority over that specific layer. Without labels, the model guesses, and guesses are unreliable. With labels, you get consistent, directed output.
On Floyo, you can access Uni-1 through Luma AI's ComfyUI API nodes. Floyo handles the API connection and GPU, so you can start generating and editing without managing infrastructure.
What can you create with Uni-1?
Uni-1 handles text-to-image generation, image editing, multi-reference composition, style transfer across 76+ styles, character consistency, sketch-to-image, identity and pose transfer, temporal aging sequences, and multi-turn iterative editing. All tasks use the same Create or Modify modes with role-labeled reference images for precise control.
| Capability | What It Does | Use Case |
|---|---|---|
| Create Mode | Generate new images from text. The model reasons through spatial layout, lighting, and composition before rendering. | Concept art, product photography, cinematic scenes |
| Modify Mode | Edit existing images with surgical precision. Change time of day, add elements, adjust lighting while preserving structure. | Photo retouching, scene adjustments, client revisions |
| Multi-Reference | Provide up to 9 images, each labeled with a role (style, character, composition, lighting, texture, mood, color palette). | Directed generation, brand consistency, complex scenes |
| Character Consistency | Use a canonical reference image as a CHARACTER anchor. Maintains consistent appearance across scenes. | Storytelling, character sheets, social campaigns |
| Text Rendering | Generates accurate, legible text in images. Supports English and Chinese, including signs, labels, and infographics. | Marketing assets, banners, comics, educational content |
| Iterative Refinement | Lock a seed, then change one variable per generation. Create-to-Modify chains let you explore compositions, then refine details. | Design iteration, controlled exploration, art direction |
How does Uni-1 compare to other image generation models?
Uni-1 ranks #1 in human preference Elo for overall quality, style and editing, and reference-based generation. It outperforms GPT Image and Nano Banana 2 on reasoning-heavy benchmarks like RISEBench. Its per-image cost at 2048px is about 10-30% lower than Midjourney and GPT Image at comparable resolutions.
| Model | Reasoning | Reference Control | Text Rendering | Styles |
|---|---|---|---|---|
| Uni-1 | Structured internal | 9 role-labeled refs | EN + CN, near-flawless | 76+ |
| GPT Image | General multimodal | Standard | Good | Limited presets |
| Midjourney | Prompt matching | Style refs | Moderate | Many (via prompts) |
| Ideogram | Limited | Standard | Strong | Moderate |
| Stable Diffusion XL | None (diffusion) | IP-Adapter/ControlNet | Weak | LoRA-based |
Source: Luma AI official documentation, RISEBench results, and human preference evaluations as of March 2026.
How does Uni-1 work?
Uni-1 is a decoder-only autoregressive transformer that operates over a single interleaved sequence of text and image tokens. It parses your instruction, performs spatial planning, renders pixels incrementally, verifies constraints, and continues rendering. This is the same general pattern that lets language models reason through problems before producing output.
A prompt like "a glass of water next to a book on a wooden table with sunlight from the left" gets decomposed into spatial relationships and lighting logic before any image data is produced. The model reasons through where objects should sit relative to each other, how light should fall, and what the scene should look like as a whole.
Luma's research shows that training a model to generate images also improves its visual understanding. On the ODinW-13 object detection benchmark, the full Uni-1 model scored better than an understanding-only variant. This supports the idea that perception and generation strengthen each other when trained in a unified architecture.
On Floyo, Uni-1 runs through Luma AI's API nodes in ComfyUI. You can chain it with other models in the same workflow. Generate a character reference with Uni-1, then animate it with a video model like LTX 2.3 or Wan 2.7, or use Uni-1's Modify mode to iterate on outputs from other image models.
Note: Uni-1 is an image model. It does not generate video or audio. For video, pair Uni-1 with Luma's Ray models, LTX 2.3, or Wan 2.7 in a ComfyUI workflow on Floyo. Uni-1's API is rolling out in 2026. Direct API access is on a waitlist; on Floyo, you can access it through Luma's ComfyUI API nodes.
Frequently Asked Questions
Common questions about running Uni-1 on Floyo.
Uni-1 runs as an API node, so generation costs come from your API Wallet (separate from FloTime). Floyo gives $1 in free API credits on signup. After that, Uni-1 costs about $0.09 per image at 2048px resolution.
Open Floyo in your browser, find a Luma AI workflow (search "Luma" or "Uni-1" in the template library), and click Run. Floyo handles the API connections and ComfyUI environment. No local install, no Python setup required.
Luma AI, the company behind Dream Machine and Ray video models. Uni-1 was released on March 5, 2026 as Luma's first model in their Unified Intelligence research line. It is deployed to enterprise clients including Publicis Groupe and Adidas.
Create mode generates new images from scratch, optionally guided by reference images. Modify mode edits an existing image while preserving its composition and structure. Use Create when you want something new. Use Modify when you want a version of something you already have.
Uni-1 reasons through your prompt before rendering, which makes it stronger at complex scenes, spatial relationships, and structured instructions. Midjourney uses diffusion-based generation, which can produce wider aesthetic variation per attempt. Uni-1 supports up to 9 role-labeled reference images (vs. style references in Midjourney) and renders text in images more accurately. Uni-1 is about 10-30% cheaper per image at comparable resolutions.
Yes. Floyo runs ComfyUI, which lets you chain multiple models. Generate a character with Uni-1, then animate it with LTX 2.3 or Wan 2.7, add a voiceover with Step Audio EditX. All in one pipeline, all in your browser.
Yes. Uni-1 renders accurate, legible text in both English and Chinese, including signs, labels, structured infographics, and calligraphy. Most other image generators handle in-image text poorly. Uni-1's reasoning architecture plans text placement as part of its composition step.
Yes. Images generated through Uni-1 can be used for commercial purposes including advertising, client work, merchandise, and digital products. Check Luma AI's terms of service for full details.
Try Uni-1 on Floyo
Reasoning-first image generation with Create and Modify modes, 9-image referencing, and 76+ styles. Run it in your browser.
Related Reading
Character and Concept Design on Floyo
Ad Creatives for Social and Web
Last updated: April 2026. Specs from Luma AI official documentation, learning center guide, RISEBench results, and human preference evaluations.