Open Source Factory

IMAGE, VIDEO & AUDIO LTX-2.5: Reproduce the Setup Before Quoting the Speed Open Source Factory · Aug 11, 2026 · 7 min read LTX's 6.8-second result uses two GB200 accelerators for a specified 720p clip. The open weights and ComfyUI integration make other setups possible, not equally fast. For a local trial, keep the artifact, resolution, frame count and complete pipeline time in the record. In this article Architecture and released variants The 6.8-second test setup LTX's blind preference test Documented local hardware paths License conditions Earlier open-video releases World-model and real-time plans The August 11 release generates video and audio together. It offers Hugging Face weights, launch-day ComfyUI integration and a hosted API. Each path has different hardware, pricing and reproducibility conditions. That day-one ComfyUI integration is the part that actually matters. ComfyUI has quietly become the de facto prototyping environment for open generative media, and...

MODELS & COSTS MiniMax H3's 4B Encoder Swap: One 1080 Ti Experiment Open Source Factory · Aug 10, 2026 · 4 min read The encoder swap solved a concrete setup problem on this 11GB card, but the controls are the part worth inspecting. With one prompt and seed, the learned projection produced the intended scene while the zero and identity matrices did not. That is a small result, with a useful boundary. In this article The encoder being replaced The verification: "a red ball on a wood table" Timing and offloading on the 1080 Ti Observed losses and test limits What this experiment supports The encoder being replaced The baseline encoder is Qwen3-VL-32B truncated to 50 layers, with 15.7GB of NVFP4 weights producing a [seq, 5120] conditioning tensor. That exceeds the card's 11GB VRAM before accounting for the video model. It explains why encoder placement was worth testing on this setup; other machines may have a different bottleneck. ClipProj, an MIT-licensed proj...