Here are real diffusion-model options people commonly use to transform a human photo into a building / architecture (usually via image-to-image editing, not a single “human→building” model):
- Use img2img with a strong prompt like: “a skyscraper shaped like a human silhouette, architectural photography…”
- Works best when paired with structure controls (below).
- This is the “base” most workflows build on.
- You give it the input photo + an instruction like:
“Turn the person into a modern glass office building, keep the background similar.”
- It’s explicitly trained for instruction-guided edits (image → edited image). citeturn1search0
- ControlNet lets you keep pose/edges/depth/segmentation from the human photo while you “re-skin” it into architecture.
- For “human → building,” the most useful ControlNets are typically:
- Canny / Lineart (keeps outlines)
- Depth (keeps 3D-ish structure)
- Segmentation (lets you target “person” regions to replace) citeturn1search23turn1search11
- Similar idea to ControlNet: add an extra conditioning image (sketch/canny/depth/pose) to control generation.
- Often used when you want controllability but with a smaller add-on module. citeturn1search21turn1search1
- Lets you condition generation strongly on an input image prompt, often useful for keeping identity/overall composition while changing content/style toward “architecture.”
- Good when you want the output to still “feel like” the original image but transformed. citeturn1search12turn1search24
- If you want a consistent “human becomes building” look, you can fine-tune SD/SDXL with:
- a LoRA trained on examples of the transformation style, or
- DreamBooth-style personalization.
- This isn’t a single official model name; it’s a training approach widely used on top of Stable Diffusion. citeturn0search8turn0search14
For the specific effect “person literally becomes a building,” the most reliable combo in practice is:
SDXL img2img + ControlNet (canny/depth/seg) + a strong architecture prompt. citeturn1search23turn1search11
If you tell me what toolchain you’re using (ComfyUI, Automatic1111, Diffusers, Invoke), I can suggest the most direct node/model setup and example prompts for “human → building.”