SHCanada2

interesting on the tracing an image vs processing an image. When I was trying Chatgpt, it actually suggested chatgpt to create a python script to process images x,y, location of walls, windows,,etc in the image (for undimensioned, but containing a scale) and then creating a CA instruction set. ..havent tried it yet but sounded interesting in that it would be reusable

do that. There are some very specific things to know to make your life easier. and search this forum for all of the "workarounds" and tips when you have an issue. The best tip i think I ever learned was to keep walls aligned between floor levels using the alignment tool. That and creating rooms with different floor elevations on the same floor level is a PIA (for these types of plans you need to continually watch the floor and ceilings of the other levels as CA will adjust them without you noticing) and either create a layout and plan template so you dont have to keep doing things every time, or buy one from rene or others.

I'm using a local model which is not listed. I assume CA only listed the couple they may have tried it with

are you one the correct floor?

I would also like this. In the meantime, I take cross sections, and dimension them to the plate, top of fascia, top of heel, and ridge

did the as built app give you all those nice round numbers?

there are a couple threads on different technology and include discussions of it. I think valleyguy has the most accurate & efficient workflow. He uses the app to create the floor plan and then uses a laser to measure things and marks them down in an ipad app. I tested it on a few houses early on. It cant be used blindly on the examples I tried it on. The question you are left with is what walls do I have to change to make it accurate? I ama considering using it for floors not of interest. As with many of these apps, they measure the walls, and then try and line up common walls to form rooms and stich them together. And when the walls do not line up, the software makes decisions in order to make them line up. The problem is you don't know what decisions it made, and if they were correct. For the $1200 for the pro, i'm starting to think a cheaper lidar point cloud scanner is the better option @ 5k or less and not have to wonder which walls might be out

Thats why I like to go to site ...the robots are not here yet

that of course begs the question, cant it go watch the training videos and read the doc :o) ..and of course read all the posts here for all the workarounds... or is it pattern based training, like here MR AI is a picture of the floor plan, and here is the corresponding plan in CA I created. repeat for 100. ...but I think that might be more difficult...as it still would not "know" how to create it. in watching the revit AI video from the point cloud, my guess is that is what they did, as it still does not do it all in the output that I saw ..or maybe it is both...

thats what i do today, but it is all manual work. A simple house can be done in a couple hours. but why not have AI do it, and go do something else for a few hours. Compared to what you are doing in the video, which is instruction after instruction, based on what you want, telling it to generate a model from an image would be more deterministic ( i shouldn't have to tell it anything after the initial instruction). Maybe harder for AI, but if it sat there all day trying different things until it got it right (assuming it didnt start using roof planes for slabs) then I can go do something else. I saw a demo of a company trying to turn point clouds into revit models yesterday, but that is a different approach. The approach in your video at the start is the push button approach, which interestingly may work better, who knows. It certainly piqued my interest

what did you give it to create the initial floor plan? and the floor plan is in CA?

i had the same question you had at the end...why is it using dialog boxes, vs directly doing things. ...looks like the answer is only certain things are available to it as a tool from the MCP server. What would be interesting is if one gave it an image of a floor plan and elevations, could one tell it to build the house in CA, go mow the lawn, go get groceries, come back 5 hours later and it would be done?

interesting, what do you measure for the site plan? For me, I have no idea where the property lines are when I am onsite, so there is not a lot to measure per se. I just take pictures of the building in order to try and create representative elevations. the one thing I might measure is the main floor to grade, front and back to get the terrain accurate for the elevations. Occasionally I will take a laser and measure elevation changes between certain points if for instance they are putting in a new garage on a sloped terrain

interesting thats a lot, today I take a video, but that even misses things, and well I cannot measure from it. Curious, have you looked at handheld scanners (no tripod to setup) which produce both point clouds and 3DGS, and/or 3DGS only cameras (which are also measurable). And if so, what got you to the LX1 vs a scanner or a 3DGS camera? example 3DGS XGRIDS LCC the measurement is nice as it shows you how far off the axis you are

that is surprising. even lower end commerical grade scanners are only accurate to +-0.75 inch. higher end to 1/8" I think I have convinced myself to go the SLAM (mobile) scanner route, just for the fact that if I missed something, like dropped beam, or stairs proud of a wall, I can go back to the point cloud and measure. Plus, they all seem to have actual cameras on them, so you also get a 3d model which actually look like the interior (similar to matterport). and they can do exterior so I could definitively know building height for instance. one time HW cost, no subscriptions, you own the data, can measure everything from the desktop. ...or at least I've convinced myself. most scanner vendor software can give you a floorplan (or alternatively trace the walls) The only question left is which one to buy. I'm currently in price creep mode