OPEN_SOURCE ↗
REDDIT · REDDIT// 2h agoOPENSOURCE RELEASE
useknockout v0.6.0 expands image-ops API stack
useknockout v0.6.0 packages background removal, upscaling, face restoration, and background replacement into a single FastAPI service. It ships as an MIT open-source stack that deploys to Modal in one command, with a free hosted beta endpoint and MIT SDKs.
// ANALYSIS
The interesting part here is not just the model list, it’s the packaging: one self-hostable API that turns several strong vision models into a production-shaped service.
- –`/remove` combines BiRefNet with pymatting refinement, which makes this more than a thin model wrapper
- –`/upscale` gives teams a choice between Swin2SR and Real-ESRGAN, which is the right kind of tradeoff knob for different image types
- –Baking weights into the Docker image and targeting Modal lowers the operational overhead a lot for teams that want GPU inference without building infra from scratch
- –The MIT license across the repo and SDKs matters here: it’s positioned as an open alternative to commercial tools like remove.bg and Topaz, not just a demo
- –The free hosted beta gives it an on-ramp, but the real value is self-hosting for teams that care about cost, control, or data locality
// TAGS
useknockoutapisdkopen-sourceself-hostedvisiongpu
DISCOVERED
2h ago
2026-05-07
PUBLISHED
6h ago
2026-05-07
RELEVANCE
8/ 10
AUTHOR
KingOfAllContent