FAL Releases 3DREAL Strong v2: Open-Source LoRA Converts 3D Blockouts to Photorealistic Video

FAL has released 3DREAL Strong v2, an open-source LoRA adapter for the LTX 2.3 video model that translates grey-box 3D blockouts into photorealistic footage while preserving exact camera paths, composition, and timing. The new version adds multi-scene coherence across cuts, synced audio generation, improved talking characters, and sharper overall output. Weights for all three adapter tiers — Light, Strong, and Strong v2 — are available on Hugging Face, with a hosted `render-to-real` endpoint already running the latest version.

The release doubles as a detailed production post-mortem for a 1930s-era action short — a truck racing and jumping over a steam freight train — built entirely through a 3D-to-AI pipeline. The workflow splits responsibilities cleanly: Blender (driven via the Blender MCP without hand modeling) owns geometry, cameras, motion, and cuts; image models supply reference looks and per-shot keyframes; and the 3DREAL LoRA renders the final pixels. FAL documents what worked, what failed, and the full iteration count, framing the pipeline as a solution for productions that need total directorial control over action and camera — something current image-to-video and reference-to-video models don't consistently deliver.