How to Turn an Image Into a 3D World with FLUX.2

FLUX.2 reads literal compositional prompts. Write the source scene with explicit depth and composition language: "Wide shot of a Tokyo backstreet at dusk, foreground left: a vending machine glowing pink, mid-ground center: wet cobblestone alley with neon reflections, background right: small ramen shop with warm interior light, atmospheric depth, photorealistic, 4K." The literal staging gives the world node clear depth cues to reconstruct.

FLUX.2 Pro's prompt adherence is meaningfully tighter on detailed compositions. Pin Pro on the source reference for any Image-to-3D-World workflow — the prompt-literal staging gives the world node more accurate depth signals. Base tier still works for exploratory drafts; switch to Pro before the final source generation.

Save the scene-style language in a Text node: "Photorealistic cinematic photography, atmospheric depth, soft natural light, anamorphic compression, sharp foreground with bokeh on the far-field." Wire it into the FLUX.2 source node as a brand-style prefix. This keeps the look consistent if you generate multiple source candidates before picking one to feed the world node.

Drop a World Labs or Image-to-3D-World node onto the canvas and wire the FLUX.2 output as the input. Generation takes around 5 minutes for a full navigable world. The output is a canvas-internal scene preview — you can orbit, pan, and screenshot inside the canvas, but cannot export the world as a portable mesh or splat file from Astorie.

Inside the navigable preview, capture stills from four angles: front view, three-quarter left, three-quarter right, back/over-shoulder. Each capture lands as an image node on the canvas. Capture more angles than you think you need — re-running the world node produces a different scene, so screenshot first, iterate later. These are the shot starting frames for downstream video.

Wire each captured still into its own Sora 2 or Kling 3 video node — image-to-video with the captured angle as the starting frame. Add cinematographic motion prompts. Each video clip inherits the world; only the camera move changes. The locked location is what makes the multi-shot sequence read as one place across cuts.