LTX Studio 2.3 GGUF 12GB adds multi-image workflows for faster video
New chunk feed forward cuts VRAM usage while supporting 4-image input workflows...
LTX Studio has rolled out version 2.3 GGUF 12GB, a significant update to their AI video generation platform. The headline feature is the addition of multi-image input workflows for first-frame/last-frame (FFLF) generation, supporting up to 4 input images pre-configured and ready to use. This allows creators to define both the start and end frames of a video sequence, with two additional input slots available for further customization. The update is available on CivitAI for download.
Under the hood, LTX 2.3 introduces several performance optimizations. Chunk feed forward reduces VRAM usage, making the model more accessible for users with 12GB GPUs. The update also includes a new upscaler v1.1 and distill LoRA v1.1, along with rewired nodes and removal of unnecessary components. Previews have been restored, and the IC Detailer LoRA is now applied only to stage 2 of two-stage workflows (excluding v2v). Users are advised to consider V1.0 workflows obsolete and adopt these new versions as the default.
- Multi-image input workflow supports 4 pre-configured images for FFLF generation
- Chunk feed forward reduces VRAM usage for 12GB GPU users
- New upscaler v1.1 and distill LoRA v1.1 with streamlined node setup
Why It Matters
Gives video creators more control over AI-generated sequences with multi-frame input and lower hardware requirements.