- Featureblog.fal.ai
Ulysses Unbound: Experiments in Communication–Computation Overlap
The article discusses advancements in the Ulysses communication-computation overlap approach for video diffusion models. The core idea is to use context parallelism with sharded heads, which maps well…
- blog.fal.ai
Instruction-level control with Inline Elementwise ASM in Triton
Triton, a Domain-Specific Language (DSL) for writing fast GPU kernels in Python, has introduced a new feature called inline elementwise assembly. This feature allows developers to inject device-specif…

- Launchblog.fal.ai
Kling 3.0 Prompting Guide
Kling 3.0, a new model designed to understand cinematic intent, has been released. This version introduces significant improvements in multi-shot generation, supporting storyboards of up to six shots …
- blog.fal.ai
Crafting Efficient Kernels with Epilogue Fusion
The article discusses the concept of epilogue fusion in the context of GEMM (General Matrix Multiply) operations in machine learning workloads. Epilogue fusion is a technique that eliminates intermedi…
- Launchblog.fal.ai
Kling 3.0 is Now Available on fal
Fal.ai has released Kling 3.0, a state-of-the-art generation stack for video and image creation, focusing on structured storytelling. This update introduces shot-by-shot direction, consistent characte…

- blog.fal.ai
Grok Imagine is Now Available on fal
Fal.ai has introduced Grok Imagine, a new multimodal release featuring five new model endpoints that support both generation and editing across image and video workflows. This release enables teams to…

- Featureblog.fal.ai
Chasing 6+ TB/s: an MXFP8 quantizer on Blackwell
Fal.ai has developed an MXFP8 quantizer in CuTeDSL that achieves over 6 TB/s on NVIDIA's Blackwell B200 GPUs. This quantizer writes scale factors directly into the packed layout expected by Blackwell'…

- Launchblog.fal.ai
LTX 2.0 is Now Available on fal
fal.ai has launched LTX 2.0, a state-of-the-art open-source model for text-to-video and image-to-video generation. This update introduces next-level cinematic control, synchronized audio, and advanced…

- Launchblog.fal.ai
Chatterbox Turbo is now available on fal
Chatterbox Turbo, an open-source, ultra-fast text-to-speech model, is now available on fal.ai. This model is designed for real-time voice AI, offering sub-150 ms time to first sound, expressive parali…
- blog.fal.ai
Seedream 5.0 Lite Prompting Guide
Seedream 5.0 Lite, released in February 2026, introduces significant advancements in image generation capabilities. This guide is based on extensive testing and highlights the model's unique features …

Track Fal on autopilot
- · Weekly AI brief: narrative summary of what shipped, every Monday 9 AM
- · Email or Slack alerts, or chat with the archive in your dashboard
- · Add Fal + up to 2 more competitors free, no credit card