Show HN: I made an app that uses AI to change your interior

This is really cool! My #1 feature request would be to try to preserve the floorplan and not change any of the walls, windows, or doors. That sounds like a very difficult challenge though. I&#x27;ve seen some new AI tools recently that can render 3d objects. It would also be awesome if I could upload a 2D floorplan to get some ideas from a top-down view.<p>I actually tried this myself with DALLE-2 recently, but it had very poor results. I&#x27;m atrocious at interior design and constantly make mistakes and bad choices, so I regularly look at subreddits like &#x2F;r&#x2F;designmyroom and &#x2F;r&#x2F;malelivingspace. I&#x27;m looking forward to seeing some good tools in this space. I can imagine a polished product that is actually of the room layout. Imagine if it could even scrape second-hand furniture photos from ebay and TradeMe (in New Zealand) and automatically figure out the size and generate 3D models that you could place in your room (e.g. using SweetHome3D.) That would be extremely useful.

Feature request: use img2prompt to allow people to request things like &quot;change the color of the couch to blue&quot;<p><a href="https:&#x2F;&#x2F;www.reddit.com&#x2F;r&#x2F;StableDiffusion&#x2F;comments&#x2F;xboy90&#x2F;a_better_way_of_doing_img2img_by_finding_the&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.reddit.com&#x2F;r&#x2F;StableDiffusion&#x2F;comments&#x2F;xboy90&#x2F;a_b...</a>

It&#x27;s a very clever idea for an awesome, interactive app but in practice the results are extremely terrible. The output has almost nothing to do with the input, the outputs are all really just slight variations on what was clearly a very small training set, and the results are hardly even &quot;possible&quot; or &quot;sane&quot; suggestions with extremely obvious AI artifacts like impossible shapes or walls that are also chairs that are also windows that are also some weird thing no one can name.

Have a chat with IKEA - if you could integrate this with their inventory and then link to the specific items shown in generated images (for a revenue share or some other arrangement) you could have a slam dunk on your hands.<p>They&#x27;re probably the only retailer with a big enough range and significant enough global presence for it to work well for users and scale.

I&#x27;m building a PDF generation service (DocSpring), so it&#x27;s probably time to look into using AI to detect fields on uploaded PDFs so that it&#x27;s easy to create PDF templates. That&#x27;s been in my backlog for years. It was probably easy to build this many years ago, but maybe it&#x27;s even easier now with the latest AI libraries and services.<p>I should really start to get familiar with AI and learn how to use things like PyTorch. I&#x27;m a bit nervous though, because it feels very different to all the programming that I&#x27;ve been doing for over a decade. I&#x27;m using to building CRUD web apps and mobile apps, writing SQL queries, setting up servers on AWS, etc. I&#x27;m not very good at math and I don&#x27;t really understand how neural networks work or which kind of algorithm I should choose, so it feels like learning a whole set of new skills from scratch. But I can probably figure out how to use some off-the-shelf libraries and frameworks and string something together. Maybe this will be my next weekend project.

I gave it a try. Took a picture of my bedroom, uploaded it at interiorai.com and i got very amazing interior design ideas.<p>I don&#x27;t have money to afford such beautiful furnitures and accessories but still it is so amazing haha

This is awesome, but I&#x27;m a bit concerned about the &quot;latest renders&quot; feature potentially doxxing people. Are you checking for identifiable information somehow?

i find it really cool that you can drop $1k a day on a project like this: <a href="https:&#x2F;&#x2F;twitter.com&#x2F;levelsio&#x2F;status&#x2F;1577006975270866944?s=20&amp;t=J_mnvmg9JrQl5NyDrNyr3w" rel="nofollow">https:&#x2F;&#x2F;twitter.com&#x2F;levelsio&#x2F;status&#x2F;1577006975270866944?s=20...</a><p>this is the kind of aggressiveness&#x2F;risk taking that people need to find hits in the new AI business

This is really great. My only suggestion for improvement is to enable the web UI to allow pasting directly from clipboard.<p>Can I ask what&#x27;s going on under the hood? My guess is you&#x27;re using Stable Diffusion with the uploaded picture and a phrase like &quot;in a modern style&quot; for a prompt?

This is just awesome. I prompted Midjourney and Dall-e 2 ealier and the outcome was rather amusing.

I don&#x27;t see any licensing information for the generated image. Even the signup modal has no link for &quot;our TOS&quot; that I must agree to. Am I missing it?

I tried with my home office and the ideas are amazing

Would love to understand what it uses behind the scenes?

This is awesome