Image-to-Image
Text-to-Image
Sketch-to-Image
CAD-to-Image
For CAD renders, ControlNet’s Depth Maps are the best method of producing reasonably accurate renders, in a matter of seconds.
This can be combined with Canny ControlNet and Lora’s for optimal results.
All of the images below were produced using the depth map shown, with the following prompt and settings:
Prompt: beautiful suburnban home in the summertime, architecture render with fluffy clouds
Settings: Steps: 20, Sampler: Euler A Turbo, CFG scale: 9, Size: 1920x1024, ControlNet 0: "Module: depth_anything, Model: diffusers_xl_depth_full [2f51180b], Weight: 0.7,
Lora: HH-arch_rowhouse_XL
Lora: HH-arch_rowhouse_XL
Lora: JJsRowHouse_XL
Lora: 222ArchDrafting_XL
Lora: Flat-XL
Lora: Architect_XL
Light Renovations
Depth-to-image also allows for re-rendering existing spaces using a text prompt. This is ideal for rapid visualisation and conceptualisation of spaces.
LoRA (Low Rank Adaptation) are models that change the conteext/behaviour of larger models. These are useful in architectural visualisation for style transfer, as seen below.
Villa_F_by_Hornung_and_Jacobi_Architecture_SDXL-000008
222ArchDrafting_XL
ArchDrawing_XL
architect_XL
ArchSketch_XL
HandSketcher-XL
JJsArchitectureSketch_XL
JJsSitePlan_XL
JJsRowHouse_XL
JJsFactory_XL
JJsConcreteArchitecture_XL
JJsArchiviz_Birdview_XL
JJsApartment_XL
HH-arch_rowhouse_XL
Depth to image is even capable of handling rougher sketches and turning them into image sketches or even full renders with loose guidance.
Sketch by Amin Zakaria
Sketch by Amin Zakaria
The better the initial sketch, the better the generated images will come out, and the more accurate to the original design you can make them.
… but that’s no reason not to get the kids involved
Sketch by Zakaria Jnr
