About the job

BetterGroup (the team behind BetterPic and BetterStudio) is a profitable AI startup scaling fast, millions in revenue, millions of requests, and increasing GPU workloads across multiple products. On BetterPic, our AI makes professional headshots accessible to anyone, fast, affordable, and 4K studio-quality. Now we’re expanding into B2B, building high-impact workflows for teams and partners. On BetterStudio we’re solving photoshoots for fashion ecommerce.

We’re looking for a Lead ML Engineer to take full ownership of our training and inference stack. If you live and breathe generative models, enjoy pushing training quality and speed, and know how to optimize real-world ML pipelines, keep reading.

This isn’t a research lab. We ship daily and care about results.

We value your time, here’s what matters…

Team Perks

Meet your Team

All headshots generated using Betterpic ❤️

About you

Your Impact

• Own end-to-end training pipelines for custom LoRAs, DreamBooth, and other checkpoint finetunes

• Architect efficient pre-training setups to maximize likeness and minimize cost

• Build, optimize, and debug image generation pipelines (ComfyUI or custom)

• Manage quality control across generations using similarity models and QA tools

• Collaborate with the backend team to integrate inference endpoints and jobs

• Research, experiment with, and apply latest stable diffusion model techniques (Flux, SDXL, etc.)
  

Tech Challenges You’ll Tackle

• Training checkpoints at scale with limited resources

• Building robust QA systems to detect failures, duplicates, and misalignment

• Managing large dataset pipelines and label models

• Speed vs. quality tradeoffs for production inference

• Auto-finetune loops, image grading, and metadata extraction

Daily Responsibilities

• Train new LoRAs and DreamBooth models on user or partner data

• Iterate and tune hyperparameters for better likeness + stability

• Analyze model outputs, optimize prompts and conditioning

• Build/extend ComfyUI workflows and custom nodes where needed

• Maintain internal model zoo and manage versioned training runs

• Review recent papers, community workflows, and model updates

What You Need

Must-Have

• 4+ years of ML experience, at least 2 years in generative vision models

• Deep knowledge of diffusion models and image generation workflows

• Proven track record building and shipping DreamBooth or LoRA systems

• Hands-on experience with ComfyUI workflows (building, debugging, optimizing)

• Python proficiency, with batch processing and job handling skills

• Comfortable integrating with backend teams via REST APIs or local scripts

• Strong focus on training performance, quality, and stability

Bonus Points

• Experience training and using LoRAs for Flux and SDXL

• Prior work on automated quality control for AI imagery

• Familiarity with Node.js and backend job orchestration

• GPU resource management, dataset cleaning, and augmentation tools

• Contributed to open-source generative ML workflows or tooling

Our Infra & Tools

• Models: SDXL, Flux, Custom Checkpoints, LoRAs

• Frameworks: PyTorch, HuggingFace, Diffusers, ComfyUI

• Infra: AWS EC2, RunPod, Render, HuggingFace Hub, and our own decentralized BetterContainer to run your inference and training :)

• Dev Stack: Node.js, Python, Redis, PostgreSQL
  

Ops: GitHub, Slack, Docker, Asana

Application Process

1. Fill out this form

2. [Take Home] Coding Challenge (1-3 hours)

• 2.1. Small development challenge

• 2.2. Brief [5 minutes max] Loom video explaining functionality, implementation and design decisions

3. Interview with the CTO (30 minutes)

4. Interview with the rest of the management team (30 minutes)

5. Offer from BetterPic

Final Notes
This role is for someone who ships. You don’t need to publish papers, but you do need to train models that work in production.

We encourage smart usage of AI tooling, but you’ll need to own your pipeline, your metrics, and your output.

Let’s build something cutting-edge that actually works in the real world.