Deploy BiRefNet on RunPod
Budget: $10 – $30 USD
I’m looking for someone who can turn the open-source BiRefNet background-removal model into a production-ready RunPod Serverless endpoint. I will import your Git repository straight into a fresh Serverless GPU instance, so what I need from you is a self-contained, minimal Docker solution that boots instantly and never fetches weights at runtime.
What the endpoint must do
• Accept an image supplied either as a base64 string or as a direct URL
• Run BiRefNet to remove the background
• Return a base64-encoded PNG with transparency
Technical specifics
– Create handler.py, requirements.txt and Dockerfile.
– Download and cache the BiRefNet weights during the image build; no network calls when the container wakes up.
– Optimise the image for size; only essential packages should be included.
– Ensure flash-boot performance so cold starts feel instantaneous.
Deliverables
1. Git repository containing the Dockerfile, handler.py and requirements.txt.
2. Clear README explaining any environment variables and the exact payload/response structure.
3. Quick verification steps so I can run docker build, push, and see the endpoint work on RunPod without tweaks.
If you’ve previously deployed Hugging Face models or worked with RunPod’s Serverless API, this should be a straightforward job. Let me know your estimated turnaround and any questions you might have.
What the endpoint must do
• Accept an image supplied either as a base64 string or as a direct URL
• Run BiRefNet to remove the background
• Return a base64-encoded PNG with transparency
Technical specifics
– Create handler.py, requirements.txt and Dockerfile.
– Download and cache the BiRefNet weights during the image build; no network calls when the container wakes up.
– Optimise the image for size; only essential packages should be included.
– Ensure flash-boot performance so cold starts feel instantaneous.
Deliverables
1. Git repository containing the Dockerfile, handler.py and requirements.txt.
2. Clear README explaining any environment variables and the exact payload/response structure.
3. Quick verification steps so I can run docker build, push, and see the endpoint work on RunPod without tweaks.
If you’ve previously deployed Hugging Face models or worked with RunPod’s Serverless API, this should be a straightforward job. Let me know your estimated turnaround and any questions you might have.