We kicked off 2026 by hardening Pixeltable, squashing 20 edge-case bugs across 6 releases to make the core more robust. We addressed corner cases in video frame extraction, view synchronization, index serialization, aggregate functions, and cross-instance data sharing. You can find the details in the full changelog.
Beyond stability work, we shipped several new capabilities this month.
RunwayML Integration#
Pixeltable now integrates with RunwayML for AI image and video generation. You can generate images from text prompts and reference images directly within your Pixeltable workflows:
Gemini Text Embeddings#
We added support for Google's Gemini embedding models. These 3072-dimensional embeddings work seamlessly with Pixeltable's embedding index:
FP16 Embedding Indices#
Embedding indices now default to half-precision (FP16) using pgvector's halfvec type. This doubles the maximum supported dimensions from 2000 to 4000 while cutting memory usage in half—a good tradeoff for large embedding models where you need scale. If you need full precision, set precision='fp32'.
New uuid7() Function#
We added a uuid7() function that generates time-ordered UUIDs. Unlike uuid4(), which produces randomly-ordered values, UUID v7 values sort chronologically. This means new rows cluster together in database indexes, improving insert and range-query performance:
New export_sql() Function#
The new export_sql() function lets you export any Pixeltable table or query result to an external SQL database. It automatically maps Pixeltable types to the appropriate SQL types for each database dialect:
In the News#
We hosted a workshop on Vanishing Gradients about building multimodal AI systems!
What's Next?#
For the complete version-by-version breakdown, check out the docs changelog or GitHub compare view.
You can also:
- Browse the full changelog
- Explore public datasets
- Join our Discord community
- Watch tutorials on YouTube
Happy building!
— The Pixeltable Team
