Member of Technical Staff, Real‑Time Video Generation
Senior Video Production role at Cantina. Europe · full-time · €190,000–€225,000 base.
Growth Roles summary, based on the employer's posting.
What you'll do
- Run large-scale training and fine-tuning of video models across multi-GPU and multi-node setups, owning the loop and stability
- Build and maintain video data pipelines covering decoding, sampling, filtering and conditioning alignment at scale
- Shrink capable research models into fast ones through teacher-to-student distillation, step reduction and speed trade-offs
- Push models to real-time via profiling, memory tuning, quantization tactics and runtime improvements
- Wrap models into lightweight inference APIs and prototypes, working with product to ship user-facing features
What you bring
- At least two years building and shipping ML systems with real ownership
- Solid PyTorch and Python across both training and inference code
- Hands-on work training or scaling generative models, ideally video generation, not just using checkpoints
- Experience with distributed training on large runs using DDP, FSDP or DeepSpeed-style workflows
- A track record of measurable latency, memory and cost wins through profiling and optimization
Who this fits
This suits an applied research engineer who is happy spending most of their time on training, distillation and making models run fast, then bridging that into shipped product. You must be comfortable owning the full path from data to a real-time experience. The role is remote within Europe, in a timezone within two hours of GMT. Base pay runs 190,000 to 225,000 EUR per year plus bonus.
From the employer
About Cantina:
Cantina Labs is a social AI company, developing a suite of advanced real-time models that push the boundaries of expression, personality, and realism. We bring characters to life, transforming how people tell stories, connect, and create. We build and power ecosystems. Cantina, our flagship social AI platform, is just the beginning.
If you're excited about the potential AI has to shape human creativity and social interactions, join us in building the future!
About the Role:
We’re looking for a Member of Technical Staff with hands‑on experience building large‑scale video generation models—from data and training to distillation and acceleration into a fast, production‑ready model. Our models are human‑centric and product‑oriented: think interactive characters that can respond to text/audio/image inputs and generate video with very low latency.
This is an applied research + engineering role: you’ll work on training runs, data, model optimization, and the “make it fast” path that turns a capable research model into a real‑time experience.
Typical time split (roughly):
60–75% training / fine‑tuning / distillation of large video models
15–25% inference optimization (latency/memory/cost), model runtime work
10–15% prototyping + product integration (demos → shipped features)
What You’ll Do:
Train and scale video generation models: run large‑scale training/fine‑tuning on multi‑GPU (and when needed multi‑node) setups; own the training loop, stability, checkpoints, and iteration speed.
Own data for video modeling: build and improve video datasets/pipelines (decode/sampling, filtering/quality, conditioning alignment, storage formats), and keep the pipeline fast and reliable at scale.
Distill and compress big models into fast ones: teacher→student distillation, step reduction, architectural simplifications, and quality/speed trade‑offs to hit real‑time constraints.
Make models run in real time: profiling, memory optimizations, quantization-aware tactics where appropriate, kernel/runtime improvements, and practical throughput/latency wins.
Build the bridge to product: package models into simple inference APIs and prototypes; collaborate with product to turn research progress into user-facing experiences (interactive characters, conversational video).
Evaluate what matters: set up evaluation harnesses that track perceptual quality + temporal consistency + identity/character fidelity + latency/cost.
What You’ll Bring:
2+ years building and shipping ML systems (or equivalent), with clear ownership and delivery.
Strong PyTorch + Python, comfortable touching both training and inference code.
Hands‑on experience training or scaling generative models, ideally video generation (diffusion/transformers/VAEs or similar), not just using pre‑trained checkpoints.
Experience with distributed training and large runs (e.g., DDP/FSDP/DeepSpeed‑style workflows), and the practical debugging that comes with them.
Proven ability to improve performance in practice: latency/memory/cost optimizations, profiling, and shipping measurable wins.
Product mindset: can move from research ideas → robust implementation → iterating against real constraints.
Bonus Points For:
Experience with multimodal conditioning: audio‑to‑video, text+audio+image control, lip‑sync / gesture / character animation constraints.
End‑to‑end distillation experience (teacher/student design, eval strategy, failure analysis).
Familiarity with acceleration toolchains (Torch compile, Triton, TensorRT, ONNX, custom kernels) or model compression (quantization, pruning) where applicable.
Experience with real‑time streaming / WebRTC prototypes or low‑latency media delivery (helpful, but not the core of the role).
Technical Stack You’ll Work With:
ML: PyTorch (training + inference)
Models: large video generation (diffusion/transformers/VAEs), multimodal conditioning
Optimization: distillation, inference acceleration, multi‑GPU strategies
Product: rapid prototyping, lightweight inference APIs
Infra (supporting, not primary): Docker; cloud basics (AWS‑like services)
Location:
This role can be performed remotely in Europe, within GMT +/- 2 hours.
Compensation:
The anticipated annual base salary range for this role is between €190,000-€225,000, plus bonus. When determining compensation, a number of factors will be considered, including skills, experience, job scope, location, and competitive compensation market data.