I just thought the same and thought this was really some dark pattern not adding any trust. Also there is no really good examples or performance metrics.
There is stuff on GitHub that actually does similar stuff [0] [1] . We tried some pipelines a year ago for a project (understanding compliance relevant process diagrams) and it was still quite a challenge . Wonder what the state is now with all those vision llms and even if commercial how good that stuff is.
I really appreciate you calling this out, and I'm genuinely sorry for the bad experience. The wording was misleading, and we should have been much clearer.
To be transparent, when we started, the AI model costs quickly became unsustainable with purely free usage. We had to introduce a paid model to cover these costs and ensure we could keep the service running and improving. It was a difficult decision, and we clearly fumbled the communication.
We've since added a free trial with credits for every new user. I know we made a poor first impression, but I hope you'll consider giving it another try. Your feedback is incredibly valuable and helps us do better.
You are absolutely right, and we're sorry for the frustrating experience. Forcing a sign-up before you can even try the tool is poor form, and we apologize for wasting your time. As a new team, we're still learning, and this was a clear mistake on our part.
Based on this (very necessary) feedback, we have now implemented a no-login trial so you can test the tool without creating an account.
If you do choose to sign up later, we also provide some free credits. We have to cap the trial usage due to API costs, but we've tried to make it generous enough for a fair evaluation.
Thank you for the blunt feedback. We genuinely appreciate it and would be grateful if you’d consider giving it another look. We are committed to listening to our users and continuously improving.
They should have let us tried the tool without signing in, that is only natural. Why would I give you my time and my personal information because you claim you do something? It is deceptive and unfair to our time to ask me to sign in as the final step. Instantly closed. The new generation of developers should learn some manners.
The problem is that it's really easy to say 'blah blah our service can do xyz' but we know all too well that when using LLMs most often it doesn't live up to the hype (worse still you know different users will get different diagrams from the same source).
The 'trust us, give us your money, it'll work' model isn't viable given the current state-of-the-art of LLMs.
You are absolutely right, and thank you for the direct feedback. Forcing a sign-up before demonstrating value is a poor experience, and we're sorry.
It’s clear we got this part wrong initially. It was feedback exactly like yours that prompted us to fix this.
We have now added a no-login trial so you can test the tool without creating an account. We know we have to earn your trust, and this is a direct result of learning from our early users. Thank you for taking the time to help us get better.
That's a fair question, but our goal is the exact opposite. We want as many users as possible to try the service.
The challenge was that the costs associated with the initial free trials went beyond what our team could sustain. The recent changes are our attempt to fine-tune our model to find a balance.
Our goal is to offer a service that is both accessible for users to try and financially sustainable for us to operate and improve long-term.
That is a completely fair criticism, and I am truly sorry. There's no good excuse for asking for personal information before demonstrating any value—it's disrespectful to your time.
We've learned from this and have now changed the flow to allow for a sign-up-free trial.
That's a sharp but insightful comment. We're sorry that our initial design gave you that impression.
Our intention was never to ignore "manners" for utility, but we see now how a forced sign-up does exactly that. We made a mistake.
We are committed to improving the user experience, and based on this feedback, we have now implemented a no-login trial. We'd welcome you to try it and see that we're serious about getting this right.
They didn't even let my try it, after signing in. I hit the "insufficient credits" paywall on my first upload. So, having folks sign in served no purpose at all. Just an extra layer of customer-repellant.
That is a terrible experience, and we are sincerely sorry. You are completely right—having you sign up only to immediately hit a paywall is unacceptable, and we deeply apologize for wasting your time.
Your feedback highlighted a major flaw in our onboarding, and we've worked to fix it.
We have now introduced two changes:
You can try the product without logging in at all.
For users who do sign up, we now grant 5 free credits to ensure you can properly test the functionality.
We know we made a very poor first impression, but we hope this shows we are listening. If you're willing, we'd be grateful if you’d give it another chance.
That's a very fair question. You're right—our initial approach wasn't balanced. We're still a new team and have been learning in public as we try to find the right model.
We believe we've found a better, more reasonable middle ground now, based on all the user feedback.
We now offer a no-login trial for a quick test, plus 5 free credits for anyone who signs up to try it more seriously. We hope this feels less like an extreme and more like a fair compromise. If you're interested, we'd welcome you to try it.
We're very sorry that our actions came across as deceiving. That was never our goal, but we understand that the impact is more important than the intent.
To be transparent, we were struggling to manage the high costs of the AI models and made a clumsy decision to put a wall up too early. It was the wrong choice, and we apologize.
We've listened to the feedback and have since corrected our mistake. You can now try the product without signing in, and if you do register, you'll get 5 free credits to test it properly. We hope this action demonstrates our true intention, which is to build a product people genuinely like and trust. We'd be grateful if you'd consider trying it again.
I'd really like to find an option which can preferably be version controlled and doesn't require hard-to-remember schema (ex. plantUML).
At work it's always tough to find something which works, and which is free or already licensed (no chance to get new licenses), and which is easy enough for teammates of varying technical abilities to contribute to.
For Arch Diagrams, most people seem to jump to Draw.IO, which is nice, but I'm not sure how easily it can be version controlled (although I haven't tried). At work it usually falls into the "did you put your latest version on SharePoint" black-hole (we don't pay for the cloud syncing version of draw.io). I wanted to try Figma, since it's at least a bit more collaborative, but there aren't any good first-party templates, so maybe it's not the right place, either.
For DFDs, I'd like to try Mermaid, or D2, or PlantUML (scared by the syntax on that one, though). I've not tried any of these, right now we usually do these in draw.io too, but I feel like code-defined ones would be an easier to maintain option and can live in a repo easier.
Sequence Diagrams are currently usually done using the sequencediagram.org engine, which I'm not a huge fan of, but at least it's relatively easily handled text. I don't think there was a good VS Code integration last time I checked (I think it was some web emulator, not a built-in engine?).
ERDs, I'd also like to find a good local tool to probably just use SQL on the backend, so that it's one less conversion. I'm open to all suggestions for that, though.
I've been doing diagrams with Mermaid. The beauty is that I use Gemini to ask it to add stuff to the diagram. And there are a bunch of plug-ins for Google docs to visualize the diagrams. I just wish Confluence had a free way to do it.
tldraw is also something very easy to use but surprisingly powerful and I know they are compared a lot to Excalidraw, not so sure about draw.io - but could be worth adding to your list. It is a very extendable library with some fun genai hooks like makereal.tldraw
There’s also a “include a copy of my diagram” feature when exporting an image from drawio, which partially overlaps with the use case of this app (assuming the export in question was done with this feature enabled).
Youtube videos that are lectures with slide shows .. or PDF slide decks ..can also be a input / starting point..with some additional detection and parsing. Both can have multiple images in them.
This is an excellent idea, thank you so much for suggesting it. You're right, parsing lecture videos and PDFs to extract diagrams would be a powerful feature.
We've officially added it to our backlog. We're excited about this one and will prioritize it.
We genuinely appreciate you taking the time to share your thoughts. Please keep the great ideas coming!
Draw.io already allows creating editable PNGs and SVGs: it embeds the diagram definition in the image metadata. You can then import this image like you would import any other .drawio diagram
The VSCode integration is the cherry on top. All files with the correct extension (such as xxx.drawio.png) will automatically open for edit in the draw.io UI embedded in the IDE.
I use this feature often to build on top of the previously made diagrams.
This is a very well done site but perhaps a bit too much of a demand test to be on HN which is extremely tech heavy (the typical reader is likely to wonder "I can already ask an LLM to do this. Why would I pay this company to do this for me?"). This would do very well on other, less tech heavy sites. I would also suggest some changes.
Above the fold, there's a lot of pitch how Draw.io requires "No credit card, no trial periods, no BS." and "It's genuinely free - has been for years, always will be. That's why millions use it, and why converting your images to Draw.io format makes so much sense. Your converted files will always be editable, no subscription required"
Below the fold and at the very bottom, this service itself starts off at $5/mo on sale right now.
I would imagine this would confuse people. They might interpret that this service is free too, then be suprised at the "no free tier" and some are going to be outright angry and very vocal about it.
So I would change the messaging along the lines of "We help you convert images into Draw.io so you pay us just once for the diagram you want converted"
To enhance the message, I would further say "This is how much we sponsor Draw.io for enabling our own business" and write blog posts about the struggles to build the service or even open source the methods to fine-tune a model to do the conversion.
Wow. Thank you for such a thoughtful, detailed, and genuinely helpful comment.
Honestly, after a day of fielding a lot of (justified) criticism, reading something so constructive and encouraging is incredibly moving for our team. We're trying to create a good experience for everyone, and it's clear we've made some major mistakes.
You are absolutely right about the core problem in our messaging. In our effort to praise Draw.io, we completely missed that we were setting a misleading expectation for our own service. Your suggestions on how to fix this—clarifying the value proposition, sponsoring Draw.io, and sharing our journey—are all fantastic ideas. They're not just tweaks; they're a roadmap for how we can build trust and communicate better.
We'll be discussing your advice as a team and will use it to guide our next steps. Thank you again for your generosity and guidance. It is immensely valuable.
That wasn't a deliberate choice but a bug in our front end. We've just pushed a fix for it. The showcase examples should be visible to everyone now without an account.
