keenthinker

Hi, my wife's request over the weekend to create a tool for her and her business partner turned into a new little product. With this app you can create realistic team and group photos using just simple selfies, passport photos, image croppings, and similar inputs. Currently, up to 8 people can be added to a single image, and the result looks great. The latest AI image models from Google and OpenAI - "Nano Banana" and "Image2" - are available for the image generation. What sets this app apart from many others is that the result is available instantly. I'm currently working on increasing the number of people while maintaining the quality of their appearance. What do you think - hit or a flop? :-)

Hi everyone, Pavel here - the maker.

A few months ago AI imaging became a trend and everyone was creating prompts ... using text and in English.

I thought to myself - couldn't I use AI to translate a prompt from the language of my choice into English and create an image from it? And then it occurred to me that it would be even better if I didn't have to type out my prompt, but could speak it in my preferred language and create an image at the end (again, using AI for audio transcription, of course).

And so TextUnbox Voice Drawing was born!

The service is built up on three AI layers: - the first layer does the speech recognition (transcribe audio to text) - the second layer translates the recognized text into English (planning to make this optional, since OpenAI understands a lot of languages too) - finally the third layer generates an image from the "prompt" using OpenAI DALL.E engine (currently working on a fourth layer to enhance the prompt automatically using custom trained AI)

I'm giving 25 "Textunbox 100" licenses for free for everyone who wants to try out the app. Use the following discount code: showhn-y23

Feedback is much appreciated and let me know if you have any questions.

Thank you! A good and understandable question. :-)

The API is deployed from GitHub directly (main branch), so the code in GitHub is what is online. There is no logging, no tracking, nothing at all. The idea is to build trust with the users, because everyone can check that nothing is logged. I hope I can do that! :)

I am also open anytime for security reviews of the vercel project if desired (there are no log sinks configured, not even for statistics).

Last point: even if some web application uses the REST API to generate phrases, IMO it is impossible to correlate web address (ip), username and password/uuid/wordlist, even if something is logged (the only thing that could be logged would be IP address). Even if you could get the address of the service from the ip (for example nslookup reveals, that for service.domain.com the phrase "passWORD1!" was generated on 28. july 2020), you need to know for what purpose and/or functionality (login, log, whatever) the phrase was created. Assumming it was for login, you need to check all username + this password combinations, and username is unknown, or all reset password, or ... The query to generate the password containts only generation parameters and no further details. Hope this answer makes sense and helps :)

Thank you for your feedback.

I can understand your point of view and I do believe, that it will take time, to trust online services that deliver such sensitive data as passwords and ids. That is why it is open source - to be transparent from the beginning. Nevertheless it seems to me that more an more (micro) services will emerge and replace a lot of the "classic" software that we are used to and accustomed to have offline.

Regarding the "offline uuid" - currently working on creating a npm package for offline usage. Will be available soon.

Regarding the Content-Type: it is set currently automatically to "text/html" when generating only one item (no matter if it is password, uuid or word list) and to "application/json" when generating more than one item. The idea was that the default option for each endpoint (one item, default settings) should deliver a result that can be used/copied immediately and not be formatted, if the api was called in the browser. I guess it would be fine to set text/plain when delivering one item. Thanks for the hint.

Hi everyone, Pavel here - the maker. I love to automate repetative tasks and standardise processes with the help of software. I do believe that creation of strong phrases, ids, etc. should be secure, transparent and reliable. That is why I build Hipstapas - to have online single source of strong phrases (which I already do use - dogfooding :-). Use cases that I can think of: - resetting/rotating passwords - creating secrets, that only the client should now - password managers and the like - adaptation for internal us

The service itself is a REST API build with pure JavaScript (on top of node.js). It concists currently of 3 endpoints (FAAS) hosted on vercel (formerly zeit). There is also an interactive playground build with Bootstrap/HTML/JavaScript (currently only for passwords, uuid and word lists are comming). Custom policy features is also on its way - this could be something that could be probably a paid service if the concept is at all of interest.

Let me know if you have any questions. Any feedback is much appreciated!