Do you mean Flux 2? Unfortunately, I don't use it, but if you're interested, I can tell you how I created the training dataset for this model. I went to https://chat.qwen.ai/ > Create Image, and by attaching each reference image to the following prompt, it transformed them into the correct images. Unfortunately it only allows you to make 20 free images per day, so either you subscribe, or create multiple accounts. This is the prompt:

"Transform the provided input image into a precise 3D wireframe visualization in the style of a blueprint or architectural rendering. Use only thin, white, continuous lines to outline all edges, contours, and structural details of the main subject and any surrounding elements from the input image, such as buildings, trees, streets, or objects. Do not fill in any surfaces, colors, textures, or shading—keep it strictly as a hollow wireframe mesh with no solid areas. The background must be a solid gradient sky blue, starting from a deeper blue at the top (#0099FF) fading subtly to a lighter blue at the bottom (#66CCFF), with no other elements in the sky. Maintain a low-angle perspective view from below looking upwards to emphasize height and depth, similar to an urban scene render. Ensure the wireframe lines are uniform in thickness (about 1-2 pixels wide), anti-aliased for smoothness, and densely packed to capture fine details like windows, arches, or foliage without overlapping or cluttering. Position the main subject centrally in the frame, scaling it to fill approximately 70% of the composition, with any secondary elements (like trees or poles) placed proportionally around it to match the input's layout. Avoid any text, labels, watermarks, or additional decorations. Output the result as a high-resolution image (at least 1024x1024 pixels) in PNG format for clarity and consistency across generations."