Architects' tools

And the third work from Ekaterina Zakablutskaya. Also LORA, but for SDXL (download here). This model is focused on generating terraced buildings. In the presentation that I attach below, it is interesting to read about the experiments with #revit . Overall the model is very cool with the materials and vegetation. It's probably one of the most realistic architectural SDXL LORA I've seen. Here are some images that I generated with it.

I would like to share the results of the #ai course I ran in May-June. there were 8 classes in total, between which students could do assignments and at the end do a final project. both homework and the final project were not mandatory, so not many participants (three out of seventy) decided to do a coursework, as a result of which they could either develop their own workflow in #comfyui or train the LORA model. Yulia Mashkova made a workflow where an image of a structure created in #midjourney is used as a source, and the output is a futuristic architectural environment. it was not perfect, but it was interesting, and it can definitely be developed into a full-fledged generative tool if you give a little more freedom to the controlling nodes

and there are some screenshots of the final model with different materials

I did this project a couple years ago at Coop Himmelb(l)au. the video shows the process of creating a material in #blender, which combines a procedural map and a texture generated with #midjourney. #myworkdiary #ai

Today there was a question from a student at the training course about mirrored doors in #revit . In general, they do not need to be mirrored, but if this happens, Revit has a built-in parameter that allows you to track the problem. Unfortunately this parameter is not displayed in the interface, so you need a script. You will have to write Python code in Dynamo, as there is no standard node for this task (I may be wrong). But there is in Grasshopper, where this node, like many others, is bidirectional, i.e. through it you can set a parameter (it means you can correct the situation at once) and view the existing values.

We'll (with Stepan Kukharskiy from SA lab) be hosting a webinar on the DesignMorphine platform in December, but you can register starting today: https://designmorphine.com/education/cellular-diffusions-v1-0

And this is what the model and facade details look like #myworkdiary

A modification of the script I used to model the facade of the SKA ice hockey arena in St. Petersburg. This is a project developed by Coop Himmelb(l)au

some screenshots from my regular work in the office: a few months ago we completed a BIM model and facade for the reconstruction of a building in the center of Stuttgart. the workflow for the facades is such that: first I create a kind of analytical model with curves in #rhino, then transform these curves into #revit elements through the #grasshopper + rhino.inside script. even such simple facades are almost impossible to make directly in Revit, if you need a good correspondence of elements to proper Revit categories and neat nodes. #myworkdiary

Just shared LORA for Flux, you can download it if you use this model https://civitai.com/models/693566?notOwner=true #comfyui

Yesterday I tested the interesting SDXL LORA from Cas Esbach. I loved the results especially when combined with other models, first of all with those oriented on complex geometric shapes, in this case you get a mix of minimalism and parametric style. LORA is very strong, so for the mix I used a weight around 0.2. I combined it with my own models SDXL_Geometric Minimalism and SDXL_Deconstructivism (unpublished), to which I set a weight of 0.9.

And one more workflow, where multiple LORA models are mixed. Techniques and nodes are the same as for SDXL. In these images I mixed my model with FLUX Greenification. Upscale was done separately via workflow 24_FluxDev_ImageToImage, which can also be downloaded below.

Here is the workflow, where not only LORA is used, but also there are two variants of upscale: subtle (the first image) and creative (the second image) by analogy with Midjourney. As subtle upscale turned out I don't like it much, instead of it you can use something from my old post. Creative gives great results in terms of image clarity, but are very different from the inputs, that's why it's creative.

I would recommend two LORA models: 1. FormFinder by Ismail Seleit, in the Flux version. i personally used his similar model for SDXL a lot, so this one is definitely worth a try too 2. FLUX Greenification by Artem Svetozarov. The point here is that final images generated with a lot of green around and on the building itself, and it doesn't even have to be specified in the prompts, everything works fine just by default. The model was created as a final work for the AI course that I organized recently. The generations in the post are made with this model.

Let's continue about Flux. today's topic is LORA models. the images, by the way, are the results of my model, but I'll probably post it tomorrow, as it's not available for download yet, and today I'll talk about some others models and share workflows #comfyui