[Jesus Rafael Garcia]

I’ve been training some LoRAs of architecture with SDXL model base until I found the one I wanted, specifically my own architectural style, and it works perfectly with Canny. I wanted to use it with Canny + semantic segmentation, but when I went to look for the model, I saw that there isn’t any for SDXL only for SD1.5, or I can’t find any semantic segmentation models for SDXL ControlNet with the ADE20K color system. Does it exist? What other option do I have?.
My idea is to control the silhouette with Canny, depth with depth mapping, and with segmentation, identify walls, water, vegetation, etc.
I really don´t want to train an SD1.5 model and lose quality just to use segmentation. Thanks.

Best regards.

Jesus.

[Andrew]

I’m not aware of such controlnet model.

The SDXL canny controlnet works pretty well.

You use SD1.5 segmentation controlnet to generate something close and use multiple rounds of img2img with controlnet to transform the image to an SDXL style.

[Jesus Rafael Garcia]

Thank you for the response. Yes, the SDXL Canny works great. For standard, normal camera, architectural visualization-style images, 3 out of 5 pictures are almost perfect. However, I encounter problems with aerial views; sometimes, swimming pools appear on the solarium or roof of the villas, which is amusing. Therefore, I need some form of semantic control in addition to Canny for those views. It also has to be simple enough so I can implement it into an app that can be accessed via API. This way, the architects and technical department can simply input a reference and receive a batch of images without having to use prompts.

[Author]

Posts