DreamPic.AI

Finally our app for iPhone was approved by Apple and it is now available for download and for in-app purchases in App Store. Appeared that in-app purchases requires separate review by Apple and we had our app listed for maybe a week without ability to actually purchase generated pictures. Dozen of people uploaded their photos and I think we lost at minimum several customers.

Also since hype for AI avatars has completely gone I decided to change website pitch from just fun avatars to a tool that helps create your personal brand online by providing custom pictures with you. Also going to make accent to custom styles created by users vs pre-built styles. If you want to be unique you definitely need to have 100% custom pictures and when everyone makes pictures in same styles they are not that cool anymore. So in next update I'm going to do a lot of improvements to my custom style generator. It is based on GPT-3, actually a style is created by GPT with little modifications to produce good quality pictures. We've made another couple of sales and everyone bought custom style package.

So stay tuned for updates. And here is the link to iOS app if you want to try: https://apps.apple.com/am/app/dreampic/id6444743139 it was build by my friend in a couple of weeks :)

Continue to play with Stable Diffusion, this time I tried to do animations and got quite interesting results. Asked my friends to review if they are worth to be offered as a product and got conflicting opinions. Some think it's cool, others it's waste. So I decided to put it to the website in manual mode. If there will be a demand for the feature I will automate the whole workflow, for now I can do orders manually. Some examples could be find at the website or in this tweet https://twitter.com/DreamPicAI/status/1606311018442924032
What do you think this is an interesting feature or not?

Roughly one month ago I opened the website to public but never stopped to develop and improve its features. I was not very happy with the quality of images it generated and spend a lot of time finding the right parameters to train the model. Finally I solved the puzzle and now my service produces very nice pictures. Unfortunately It requires more expensive GPUs since model doesn't fit in memory on cheaper cards. I ended up using Nvidia A100 and had to increase prices to 15$ per batch. But I think it is better to pay more but get the best possible results rather then get mediocre pictures for less price.

What have been added:

1. A lot of completely new styles
2. Sample gallery - https://dreampic.ai/samples/
3. Ability to select styles when making an order (initially random styles were applied)
4. Ability to download trained model checkpoint file (.ckpt) (so you can use it locally or at any other service)
5. Added a feature to reuse trained model and generate more pictures in the same of different styles
6. Custom styles generated with GPT - AI model behind ChatGPT (more on this below)
7. Ability for a user to delete all their data - uploaded photos, trained model and result (everything is automatically deleted in 7 days but to give users more control over their privacy I decided to add such a feature)
8. Sitemaps for google
9. Numerous backend improvements - I do everything myself, don't use any 3rd party APIs, so my backend is quite complex. It allocates a server, deploys all necessary software, downloads model and user photos, trains a model and generates pictures. Everything is automated but can fail at any stage, so I made a lot of improvements to handle unexpected situation and resolve them automatically

During this work my 1st customer has landed and made an order for 25$ :) Since I did no marketing or promotion except my initial posts on Twitter I was really surprised.

I created a new feature - DreamPic.AI has implemented a feature that allows users to generate personalized images in a variety of styles using artificial intelligence technology. This feature is powered by a language model such as GPT (Generative Pre-trained Transformer).

Language models like GPT are trained on a large dataset of text and are able to generate new text that is similar to the text that they were trained on. In the case of DreamPic.AI, the platform may be using GPT to generate descriptions of new styles for personalized images.

To use this feature, a user need to provide some input to the platform, such as a description of the desired style or a reference image. The platform then uses GPT to generate a description of the desired style based on this input. This description is then used to create a personalized image in the specified style using another AI technology - SD model trained on photos of your face.

It is important to note that the quality and accuracy of the generated images will depend on the capabilities of the artificial intelligence technology being used, as well as the input provided by the user.

The above feature description was written by ChatGPT :) I treat this feature as experimental but fun. It greatly expands the limits of the service.

I have a friend who is iOS developer and he likes the service so much that decided to develop iPhone app for it. I added mobile APIs into the backend so hopefully we will launch an app soon, will compete with Lensa :)

Next steps. Now that I am really satisfied with the quality of the results I can start telling people about my service and do some marketing. I don't really have much ideas of how to do this except for going to people in their social networks and asking to try my service. Maybe I need to try advertising on Instagram or Twitter but I don't really have any experience with producing services to end users, previously I was selling mostly to businesses. Would be really thankful to advises on how to approach this.

I also think should I implement short video clip generation with this technology or nor, and not sure should I try using StableDiffusion v2 or not. I still treat all this project as mostly educational and fun side project rather than real business, since AI Avatars hype will definitely end quite soon and who will be customers and why then this is the open question.

Thanks for reading and supporting! Follow me to get next updates on iPhone app launch and marketing results.

I was playing with Stable Diffusion when saw how other people train DreamBooth and generate fun pictures with them. Tried myself and had a so much fun that decided to build another service despite other similar services were already launched and having good sales. So, welcome DreamPic.AI - a service where you can generate pictures starting You with the help of state-of-the-art AI models.

It took me 3 weeks to implement the whole website from scratch to the public release. I didn't use any 3rd party APIs except those from Google Cloud. Frontend is rather simple - coupe of pages and one very simple form backed by React, but backend is where the fun starts. I created a custom "cloudmanager" that spins up GPU instances at Google Cloud, configures them to train a neural network, uploads a user's pictures, trains the model and generates results.

Along the way, I built a boilerplate to run my new projects with Django, React, Bootstrap, containerized dev environment and many more features I get used to over the last couple of years.

Later, I added more content pages with generated picture samples and style selector, so users can choose what style to generate images in.

Launch was 1 week ago and here are the first results:

1. I generated tons of pictures for my friends
2. Made 0 sales

The site was visited ~100 times, mostly from Twitter, since it was the only place where I initially posted about it. Despite using Twitter for more than a decade, I have only 10 followers there since I was using it mostly in read-only mode :)

Over the last couple of weeks, dozens of websites like this were launched, all do more or less similar thing, all produce similar results. The main factor of success is the reach that their creators have to the audience. But obviously to stand out, I need to bring something others don't have. What do you think it should be? Write in comments.

Nevertheless, I am having so much fun working with this project that I am going to continue and bring more mind-blowing features very soon. I learned a lot and want to learn even more.

But of course I will be really happy if someone will use it because every creator wants its creation to be recognized.

Thank you for reading this :)