The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Made a translator on local LLMs.
Stopped using DeepL lately — too many mistakes and console spam.
So in ~10 minutes I built a browser extension in Claude that connects to a local server in LM Studio and translates any text I highlight using a local LLM.
As you can see, it’s insanely fast. Maybe not always DeepL-level quality yet, but now I can just swap different models and test what works best.
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
🚀 What I Built
A smart AI pipeline that automatically translates video ads into different languages. It clones the original speaker's exact voice and syncs their mouth movements perfectly in the new language, so brands don't have to spend thousands of dollars re-shooting ads for global markets.
🛠️ The Tech Stack
ElevenLabs Flows (For the main automation canvas)
Google Gemini 2.5 Flash (To analyze the context and text)
ElevenLabs Multilingual v2 (For voice cloning)
Lip-Sync Engine (To match the mouth movements)
⚠️ Current Limitations & Challenges
Right now, the flow does not have an automatic video/audio listener feature. Because of this missing node, I had to manually copy and paste the English script into the translation node to trigger the workflow.
🔮 What’s Next?
I am already working on adding an auto-transcription (Speech-to-Text) node at the beginning. Once that's done, the pipeline will become 100% automated—you will just drop the video link, and get the translated result instantly.
Project Link: https://elevenlabs.io/app/flows/gK3BJJB4RloMQNtqRs7o
Founder and sole designer of JointLift, a real-time social accountability platform built around lightweight co-work requests between trusted contacts. Designed the core interaction model and all product flows in Figma, then built it in React Native, Expo, TypeScript, and Supabase using Claude Code. Currently in user testing. A deliberate test of how far one designer can go end to end with current AI tooling.
A production AI platform I built and run 24/7 on my own Linux VPS, several LLM-powered tools in one interface:
• Learn — an AI tutor with spaced-repetition flashcards that generates lessons, quizzes, and hints across subjects, with per-user progress tracking.
• Hub Brain — an autonomous agent that reviews system activity on a schedule and posts grounded, cited summaries (no claim without a source reference), pushed to my phone.
• Dinner — an AI agent that reads my pantry inventory and drafts complete, kitchen-ready recipes nightly via a multi-model LLM cascade (Gemini → Claude → Cerebras), delivered to Discord.
Stack: Python backend, multi-provider LLM integration (Claude, Groq, OpenRouter, Gemini), SQLite, web push, cron scheduling, Caddy + PM2. Shipped and running in production, not prototyped.
Trending
Claude
Claude has entered the design space. How are you using Claude Design?
Contra University
Learn from expert creatives how to earn more using next-gen AI tools.
creativeaiflow
Creative AI workflows are evolving. What tools do you use, and what are their strengths and weaknesses?
portfolioreview
The best portfolios tell a story, not just show a grid. Share yours for feedback.
freelancerlife
Freelancer life is wins, pivots, and everything in between. What’s yours right now?