Rersk
Joined 5 years ago
Karma
-1
Submissions
13
Member since
Mar 8
Recent submissions
Comment · 1 year ago
Comment · 1 year ago
Thank you for your honest feedback! We’re actively improving resemblance quality while also ensuring we stay within ethical and legal boundaries. Our current goal is more about fun and stylized reinterpretations than exact lookalikes, but we appreciate your thoughts and will continue refining.
Comment · 1 year ago
Hi HN,
I'm an indie maker and just launched a small side project: Celebrity AI Image Generator – a tool that lets users generate AI images of celebrities in different styles, outfits, or fantasy settings. You can select from a preset style or enter custom prompts to create stylized portraits.
This is an experiment in combining diffusion models with prompt tuning, with a lightweight frontend. I’m trying to keep it fast, fun, and privacy-respecting (no data is stored, no login required).
Some technical details:
Backend powered by pollinations.ai with some fine-tuned models for stylistic accuracy.
Hosted on GPU-inference servers with caching.
Frontend is built in Next.js.
Image generation is done via prompt + LoRA-based style adapters.
I'd love to learn:
Does the UX make sense to you?
Are the results interesting or surprising?
What would you add or change to make this tool more useful or fun?
Any concerns from a technical or ethical standpoint?
I know the idea of AI-generated celebrity images may raise some questions around likeness rights, which I’m being cautious about. All generations are clearly synthetic, and I'm considering adding filters to avoid misuse.
If you're curious, here's the link: https://www.aicelebrity.design/ I’d really appreciate your thoughts – thanks in advance!
Comment · 1 year ago
After seeing so many Cursor tutorials, I, who has been in product operation for 5 years, finally used Cursor. My hard work paid off and I reproduced an application that generates the stories behind songs with AI. I used Deepseek API. Except that the response time was a bit slow, I was very satisfied with it.
We're sorry to hear that the result didn't resemble Jarvis Cocker as expected. Our model is still in its early stages, and while it's designed to generate stylized interpretations rather than exact lookalikes, we understand that resemblance is key to a satisfying experience.
We're actively working to improve both the accuracy and flexibility of the outputs, and your input helps us move in the right direction.