Major upgrades to the App - October 2025
Publicat pe 2025-10-03 de către Bailey Vidova
Major upgrades to the App - October 2025
For a long time now, LoveCore AI has been at the forefront of AI companionship, continually evolving to meet the needs of its users. In the past few weeks we have released the largest update to the app touching almost every aspect of it.
Text model
Originally, the LLM we have been using has been good for the text experience and it was punching above its way regarding parameters to performance. Admittedly it wasn't always the smartest.
In the couple of years since the initial release, there have been several advancements in the LLM available. Since September we switched to a much more capable model and an architecture that will allow us to scale much much easier in the future.
This new model provides a more natural and engaging conversational experience, allowing users to have more meaningful interactions with their AI companions. The responses are more context-aware, coherent, and relevant, making conversations feel more like talking to a real person.
We have also made direct improvements to allow roleplay. You can now roleplay by using *your text*
User interface
For a long time, we kept the UI simple and functional. However, the interface could be classified a little ugly. We put some effort to modernize the look and feel of the app, making it more visually appealing and more user-friendly.
We have combined the chat and the bots screens. Characters you have spoken to most recently will appear first, then characters you own
The animations for Love Roulette has been revamped to resemble cards better.
The clothes selector was redone from scratch and now characters have 4 slots for you to fill: Top, Bottom, Shoes and Accessory.
Image generation
Originally, we had to work with the primitive image models of the time and to get reasonable generation without many "bad" generations we had to put substantial scaffolding around it. That meant that the images would be good most of the time, but we couldn't follow user requests as closely as we wanted. Also, improving the system required a lot of effort.
In the meantime, several advancements on the image diffusion models have been made. We have reworked our image infrastructure, making it more simple and at the same time more powerful and performant!
While in the past we were trying to understand intent from texts and pick a "pose", we now feed the text message directly to the prompt. Just send a message like "Send me a picture of ..." and the ... will be used to generate the image, usually getting something close to what you asked. Unfortunately, that part is English only for the meantime though. So if you try to use your native language, it will still send a picture, but probably will not follow your prompt
AI Boyfriends
The new image generation and LLM upgrades, allowed us to add the ability to generate male AI companions. You can just create male characters just like female characters. They support everything that female characters support, although there is not as much content available for them yet (clothes, characters, etc).
Along with the ability to create male characters, we have also put the configration to pick which kind of characters the user is interested in.
Internationalization
AI improvements have allowed us to translate the app into 14 languages: English, French, Spanish, German, Chinese, Japanese, Korean, Italian, Portuguese, Russian, Arabic, Hindi, Greek and Romanian.
The translation is not perfect since it wasn't human generated, but we have put everything in place to be able to improve it over time.
Of course, the smarter new LLM we use under the hood, allows this to work properly and the texts can be in the users language. You can change this from your account settings if you prefer English or it is not detected automatically.
Future plans
We are not stopping here. We will keep listening to your feedback and suggestions to make the app even better. For that reason, please don't skip answering in the review prompts that pop up sometimes.
Some ideas we have are, in no particular order:
- Video generation: It is still pretty expensive to generate videos and the technology is still very young. The initial plan is to pregenerate some videos we can use for the public "official" characters.
- Voice model upgrade + special sound handling like moaning, laughing etc. The voice model we use is still the original from the first release. There is a high chance we can now do something better! In addition to that, since the voice selection is limited, it means that is feasible to also hack around put "special" sounds in
- At the moment, you can only get clothes from the Roullette. We plan on creating a store where you can buy whatever you want and paying with hearts.
- User personas. Right now user prompt is limited to just basic information. We are planning on adding more customization options for the user persona(s) so users can defined better the person they want to be in the app.
- Character memories. Characters now are limited to the context provided by their definitions and the latest messages. We plan on implementing a system where characters will remember facts about their conversations that will "survive" going out of context.
- Some sort of leaderboard for users
- Achievements, effects and other gamey features
- Improved understanding of the conversations by the system using embeddings
Try the app using the links at the footer and header
If you want to share more ideas, please reach out to at support (at) lovecoreai.com
We are excited about the future and can't wait to share more updates with you!
Publicat pe 2025-10-03 de către Bailey Vidova