| --- |
| title: Rangrez |
| emoji: 🧵 |
| colorFrom: indigo |
| colorTo: yellow |
| sdk: docker |
| app_port: 7860 |
| pinned: false |
| license: apache-2.0 |
| short_description: A self-driving dye lab on Meta models |
| --- |
| |
| # Rangrez: a self-driving dye lab |
|
|
| ▶ **[Watch the 3-minute demo](https://youtu.be/g4Z30UhPtjA)** · [A recorded live run](?replay=1) · [The 2-minute tour](?tour=1) |
|
|
| Pick a colour and press **Start campaign**. Rangrez mixes a recipe, dyes a cotton strip in its |
| simulated lab (a hidden dye lot that differs from the supplier's model), photographs it like a |
| phone would, measures it, and corrects itself until the cloth matches. Every dip photo is a |
| real-looking scene: desk lamps, daylight, tube lights, shadows, paper of unknown whiteness. |
|
|
| Without the live passcode it runs the offline policy and SAM masks cached for each scene. |
| With it, hosted **SAM 3.1** measures every photo and **Muse Wonder** runs the campaign, |
| and you can read a target colour from your own photo ("match this kurta"). |
|
|
| **Photograph your outfit:** SAM 3.1 reads its colour (a few free reads a day per visitor) and |
| Rangrez gives you a recipe to dye a matching cloth at home, in drops of blue, red and yellow |
| household colour. |
|
|
| With the passcode you can also paste the buyer's message (English or Hindi) for **Muse Spark** to |
| read into an order, run the lab **hands-free by voice** in English or Hindi (Muse Voice Transcribe), |
| and see the customer's garment in the approved shade (**Muse Image**, measured and corrected). |
|
|
| One lab, shared by everyone viewing: if a campaign is running, you are watching it. |
|
|
| Code, hardware, firmware and evaluation: https://github.com/TusharTechs/rangrez |
| Open weights: https://huggingface.co/TechTush/rangrez-glimmer-lora · Dataset: https://huggingface.co/datasets/TechTush/rangrez-lab-dips |
|
|
