Dominik from Chapter41

Formbar is a quick way to steer video generations from a single image. It automatically lifts a rough 3D clay scene from any input photo. In the studio you block out movement for characters, objects, and the camera in 3D space. One more click sends the original image plus the clay video into generation, so you get a fully steered photoreal output instead of hoping a text prompt alone gets the motion right. This prototype runs on fal and uses several models along the way: Gemini for scene understanding and prompts, Florence-2 for detection, SAM 3 for segmentation and 3D body/object lift, Marigold for depth, Nano Banana for image edits, and LTX 2.3 with the 3DREAL LoRA for the final steered video.