- Why object removal breaks on transparent surfacesThree common workarounds for removing an object from a transparent image, why each one fails, and what changed when Cleanup started treating the alpha channel as a first-class input.
- Shadows are the hard part of compositingYou cannot hand-annotate a training set for shadows, so Jasper Research rendered one, and built a single-step model that takes light direction, softness, and intensity as explicit inputs.
- Background removal is a harder problem inside a design toolFigma shipped background removal into the canvas at Config 2024, powered by the Jasper API. The bar for an embedded model is different from the bar for a standalone one.
- 2.9 billion pairs in, 105 million out: building a text-to-image model from scratchJasper Research published the full recipe for a 4B text-to-image model that renders 1024 by 1024 in under a second. The most interesting part is how much training data got thrown away.
Timing and access, answered
Jasper Image Labs models ship only after they meet the required quality and production-readiness bar, so public dates vary by capability. Register your interest to discuss the latest status and evaluation options with the Jasper team.
Yes. Early access is sales-led and depends on the model, use case, and evaluation fit. Contact Jasper to share your workflow and request access to the most relevant preview or private beta.
Jasper prioritizes customer demand, brand consistency requirements, production workflow gaps, and demonstrated model quality. Feedback from evaluation partners helps determine which capabilities move from research into supported products.
Put these models on your assets.
Early access to Labs models is sales led. One conversation covers evaluation and rollout.