UNO by ByteDance
UNO by ByteDance offers virtual try-ons, logos, character art , handles multiple subjects, styles, identities. Free to use and runs locally (NVIDIA GPU required).
Overview
ByteDance has launched UNO on Hugging Face and made it available for download to run locally—if you’ve got an NVIDIA GPU and 16GB VRAM. It’s not just another model—it’s aiming to be the model. The “One AI to rule them all” kinda vibe. Built for mixing subjects styles and identities in one go.
UNO stands for “Universal Customization from One to Many” and it’s powered by a diffusion-based system called Flux. ByteDance’s Intelligent Creation Team (yeah the folks behind TikTok) cooked this up to handle subject-driven image generation better than anything before it.
In simple words it means: You can take people or objects from different pics mash them together apply a style and boom—new consistent images.
And it does all that without needing a special syntax or markup language like with Omnigen.
Run UNO Locally
- GPU: NVIDIA only for now
- VRAM: You’ll need at least 16GB
- Disk Space: 42.55GB
- One-click install: Go to pinokio.computer and launch it using Cocktail Peanut’s setup
- UI: Gradio-based launcher
Less-to-More Learning Style
UNO uses a clever 2-step training process. First it learns to handle single subjects. Then it scales up to multiple ones. This “less-to-more” approach keeps things sharp and consistent even when you throw lots of stuff at it.
It also skips the pain of real training data by making its own using AI prompts and image mashups.
Cool Stuff You Can Do with UNO
Stylized Generation. Like giving your dog that Ghibli-esque glow or anime vibes.
Virtual Try-On. Change clothes on a model without a model trained just for that.
Subject Fusion. Mix people objects and logos into one image—like dropping your brand’s logo onto a hoodie mockup.
Identity-Preserved Gen. Keep someone’s look across scenes or styles.
Creative Storytelling. Build characters that stay consistent across scenes or frames.
What Powers It?
- Flux diffusion engine
- UnoPE (Universal Rotary Position Embedding) so it doesn’t just copy-paste
- Gradio UI so you can click around easily
Tags
Freeware Apache License 2.0 PC-based #Image & GraphicsLinks
Early impressions were mixed. A few users ran it with different prompts and said UNO didn’t always hit the mark especially with faces and backgrounds. One person compared it to caption injection and said it felt more like a fancy prompt generator than a real image transformer. But after tweaking prompts and looking closer they admitted it was surprisingly good at pulling in fine details like furniture placements and color matching even if the first glance didn’t wow them.
Other users pointed out UNO shines more with objects than people. Some found it really good at recreating clothing and hair but less consistent on faces. It also got praise for understanding input images and being able to apply transformations like blending or trying on new outfits which lines up with its advertised strengths. There's a clear divide though. Some felt it was just another LoRA file—not a real model—while others saw it as a helpful tool layered on top of Flux.
The technical debate got heated around the model’s so-called “open source” status. While it’s called open source in the post several commenters pushed back. They said it can’t be open source if you can’t train it from scratch. The current release includes the weights and inference code but not the full training setup or datasets. Some folks said this makes it more of an “open weights” release which has become common in the AI art community. Still others argued that in practice most of the community uses these tools for local gen and doesn’t care much about the pure definition.
One hot topic was ComfyUI compatibility. A bunch of users wanted to know if they could use UNO inside their ComfyUI workflows. Some said it already runs well through Gradio and should be easy to port. Others linked GitHub repos already working on nodes for it. A few testers said the VRAM usage is pretty high especially at higher resolutions like 704px. Even with high-end GPUs like the 3090 or 4090 they ran into slow gen times and crashes until community fixes started rolling out.
[ Reddit ]
woman wearing shorts while walking on the beachGenerated on April 12, 2025:
Useful Links
No additional links available for this tool.
This page was last updated on April 12, 2025 at 12:46 PM