Krea2 Deep Graph Spatial Control

This is a depth-guided spatial control workflow built with Krea2 Turbo, Depth Anything V2, Krea2ControlImageEncode, and a dedicated depth-control LoRA.

The user uploads a reference image and provides a text prompt. The workflow extracts a depth map from the reference image, encodes that depth map into a Krea2-compatible control latent, applies the depth-control LoRA to the Krea2 model, and injects the encoded depth structure through Krea2ControlApply before generation.

Unlike a normal image-to-image workflow, the reference image is mainly used as a spatial guide rather than as a direct redraw target. It provides the composition, depth relationship, subject placement, foreground-background structure, camera angle, and overall perspective. The text prompt defines the new subject, style, costume, atmosphere, and final visual direction.

The workflow also generates a comparison view showing the original reference image, the extracted depth control image, and the final generated result. This makes it useful for checking whether the output follows the intended spatial structure while still allowing creative transformation.

It is suitable for Krea2 depth control testing, spatial composition transfer, cinematic portrait redesign, poster layout control, character silhouette preservation, scene structure guidance, and depth-guided image generation experiments.