A new method (for me, of course, but I'm sure someone else has reached it before) that I'm trying at the #comfyui workshop I'm leading now: train the model on something completely non-architectural, and then mix it with architecture.
As an example, in the first class I trained the LoRA model on images of landscapes and blend it with architectectural LoRA models. The goal was to get buildings blended into landscapes in the final renders.
As an example, in the first class I trained the LoRA model on images of landscapes and blend it with architectectural LoRA models. The goal was to get buildings blended into landscapes in the final renders.
❤22👍2🙏1
Media is too big
VIEW IN TELEGRAM
The #comfyui workshop is over, it turned out to be almost completely different from last year, and after a year (definitely not before) there will be a lot of new stuff coming out and it will make sense to get together again.
I haven't written almost anything here for this month and a half, because preparation for classes takes almost all my free time (another reason why I can't afford such workshops so often), with AI it is not possible to prepare materials in advance like with all other topics, new models come out in the week between classes, and what you were going to tell is no longer relevant. That's why I chose the tactic of a TV series writer, we'll write the first season and see where the current takes us next. It felt like it worked pretty well.
I think there will be more content in the next few days, especially as there is something to tell about the workshop results. In the meantime, here's a fresh video from the last session.
I haven't written almost anything here for this month and a half, because preparation for classes takes almost all my free time (another reason why I can't afford such workshops so often), with AI it is not possible to prepare materials in advance like with all other topics, new models come out in the week between classes, and what you were going to tell is no longer relevant. That's why I chose the tactic of a TV series writer, we'll write the first season and see where the current takes us next. It felt like it worked pretty well.
I think there will be more content in the next few days, especially as there is something to tell about the workshop results. In the meantime, here's a fresh video from the last session.
❤12🔥7👍1🥰1
I wanted to post the new LoRA model and workflow for it today, but I ran out of time, so here's a preview for now (everything will be up this week). The idea is to make winter renders from summer ones. It's not absolutely perfect, it's AI after all, so a bit of Photoshop on top to keep important details of buildings.
In Photoshop it is necessary to erase the unwanted from the AI image and adjust the colors and brightness of the result.
In Photoshop it is necessary to erase the unwanted from the AI image and adjust the colors and brightness of the result.
❤5👍1👏1👌1
To make it clearer exactly how the images change, here are some gifs:
👍3❤1
by the way, you don't have to strictly repeat existing renders, you can let ControlNet go a bit and get variants
🔥8❤3
This media is not supported in your browser
VIEW IN TELEGRAM
I published the model, download here:
https://civitai.com/models/1401415
and here is an example workflow:
https://civitai.com/models/1401477
#comfyui
https://civitai.com/models/1401415
and here is an example workflow:
https://civitai.com/models/1401477
#comfyui
❤10👍5
Finally got around to testing the ChatGPT's new image features (in case you slept through the last week, you can now edit and create pretty much any image you want just using words), and it all looks pretty cool. But let's dig deeper, and see what the drawbacks are. I'll compare it to #comfyui using two examples:
1. Winter render
2. Filling an empty interior with furniture
1. Winter render
2. Filling an empty interior with furniture
The first image is an original render from a 3D model in Blender
second image - ChatGPT
third image - SDXL in ComfyUI.
It is obvious that ChatGPT is much more realistic, and it is understandable, after all, the competitor (SDXL) seems to be more than two years old.
If we talk about clear disadvantages, ChatGPT is very slow so far, generation takes 5-10 minutes, while SDXL handles the job in seconds.
second image - ChatGPT
third image - SDXL in ComfyUI.
It is obvious that ChatGPT is much more realistic, and it is understandable, after all, the competitor (SDXL) seems to be more than two years old.
If we talk about clear disadvantages, ChatGPT is very slow so far, generation takes 5-10 minutes, while SDXL handles the job in seconds.
🔥2