Losenok

Thanks for this — really useful feedback, and easy to act on. Pushed a few changes just now:

1. Removed the X from windows and doors. Delete moved to the top of the page, out of thumb range.

2. Rotation now works properly on phones.

3. Added a zoom slider (should help on the notebook).

4. New objects now pulse when placed, and the camera zooms to them if they're out of view — no more hunting for a small item, or finding it in another room.

Still working on angled walls and ceilings, and on a better experience for devices without a GPU. Those are deeper fixes and will take longer. If you get a chance to try it again, I'd like to know whether the placement flow feels different.

We scale quality off measured framerate, but clearly the floor isn't low enough on your hardware. That's a miss on our end.

If you don't mind sharing: GPU and browser? And was it slow from the first frame, or did it degrade over time?

We had a manual quality selector before we shipped auto-scaling, and pulled it on the assumption that automatic was strictly better. We will get it back to tools in the next deploy.

I do agree that hardware gap starts to shrink, similarly to what Software gap once used to be. It is much easier to avoid spending time loading the right drivers and looking up what error message on that tiny device you are working on means. Especially if AI could run in terminal itself.

I tried building a health device few years ago and got completely lost in how to setup camera and touch display with a raspberry PI. Would imagine, with AI running in a command line, it would be much easier.

Like the concept. I actually like the competitive part of it. Reminds me of playing Fifa Years ago.

I have been using Claude code and cursor daily for the past 9 months. Here is what I learnt:

1. In my experience, well-articulated prompts are the most important part. You need to tell the model exactly what and how to do to avoid hallucinations. Especially in system design, write what the end result should be, how and let the model reason and look at the existing infra first, then plan the implementation. In my experience, there is little to no coding that needs to be done after model is done implementing. Make sure to let it implement in phases, with extensive tests.

2. Model choice. It is obvious, but Claude models are the current SOTA. In my experience, Opus 4.7 extra high is the perfect balance of speed and cost-efficiency. In my experience, OpenAI models were worse in system design, but faster and better at understanding the end result. Mostly used them to verify the bigger picture. Also used composer in Cursor. Was surprised how easy it was to do web design with it.

3. Long horizon tasks. Make models build plans. Very thorough plans - for a feature or a product. It is much more aligned with a written plan.

There are more details, but this is what I noticed so far myself.

Very skeptical about the composer accuracy. I have been using it for 6 months now and it is very fast, especially compared to anthropic models, but the result it produces, especially with more difficult tasks is very shallow. It feels like it just finds the cheapest way to deliver the task.

Absolutely

We are still working on the templates. If you want to start from furnished room, you can pick a post with graphics and press "start with this design".

You can try fully designed room at https://cona.design/demo

We will make sure to add more templates tomorrow.