- Major update for the video game More than Words now available! - 4 May 2025
Major update for the video game More than Words now available: Gemma 3: Conversations are now more natural with this new language model. Response Speed: The bot's response speed has been increased. Unreal Engine 5: Lighting is now more realistic based on Lumen technology. Graphics Options: Graphics options to configure resolution and render quality. Environment: The environment has been completely redefined. Character Model: The character has been re-modeled. Animations: New idle and head movement animations. AI Navigation System: An artificial intelligence system that allows the character to navigate the environment.
- Unreal Engine 5 update on May 3 - 16 April 2025
Major update for the video game More than Words on May 3.
- AI Controller - 21 March 2025
AI system to move the character to the desired locations from a CSV file and Automatic camera tracking control. With this system, I can quickly and flexibly configure the character's movements for each level, in any type of scenario, while talking to her. + Automatic path finding system. + Automatic camera tracking system. + Player character Next step: Create the first level's scenario.
- State machine for animations - 13 February 2025
I have made progress in generating some animations and in supporting other languages. State machine for animations: idle, walk, jog, run, talk, etc. Interface translation: Spanish, German, French, Portuguese and Italian. Automatic saving of options that the player has confirmed.
- LlaMaCpp integration in Unreal Engine 5 - 31 January 2025
I have managed to integrate the LlaMaC++ project in UE5 to load Large Language Models (Meta Llama 3.1) and generate text dynamically. The main improvement is that now the speed in the generation of the responses is substantially better than previous versions. Integration of LlaMaC++ in UE5 which allows to run the LLM locally (no internet needed and anonymous conversation). The execution of the LLM supports the speeds: Normal, Medium, High and Real Time (default option, only the CPU is used). You can choose the sampler: Top K or Mirostat 2. The android has memories of past conversations. Interface to configure the android was finished. Startup interface was finished. Main interaction interface was updated. Next step: State machine for animations, localization for different languages, auto-save option, etc. Note: I have managed to apply a workaround to completely decensor the LLM, however…
- Speech Recognizer and Video Setup - 13 January 2025
Speech Recognizer and Video Setup Hi, I have managed to integrate Open AI Whisper into Unreal Engine 5 using the Whisper++ project. Also, I made a simple system that controls the video quality and resolution. Let me know what you think. For examplo, do you know of any other local speech recognizer written in C++ that could be used? Specific progress: - Integration of the Open AI whisper thread that controls the speech recognition. - Development of the system that controls the video quality and resolution. - Addition of the SETUP state in the state machine. - SETUP user interface - GRAPHICS user interface - VOICE SYNTHESIS user interface Next steps: LlaMa integration. Mainly to explore the impact of using GPUs and whether it is possible to completely remove censorship from the model.
- Text to speech and Lipsync on UE5 - 1 January 2025
I have successfully integrated Microsoft SAPI into Unreal Engine 5 to generate text-to-speech (TTS) along with my lip-syncing method. Let me know what you think. Progress in the update of More than words to Unreal Engine 5. - Main interface . - Updated Marian's 3d model - Microsoft SAPI integration in UE5 to generate text to speech . - State machine integration: Wait, Think, Speak and Listen . - State machine integration for lipsync . - Lipsync now has a more simplified algorithm with better results. - Integration of the thread that handles speaking . - Correction of a minor bug that prevented the thread from being completely destroyed. Next steps: Integrate speech recognition.
- More than words to Unreal Engine 5 - 14 December 2024
I started creating my video game More than words in UE5 from scratch. My first impressions Living coding is a headache, so turn it off. Lumen is seen well, but it reduces performance considerably, very possibly turning it off and using baked lights.
- Activation of the Mirostat algorithm and fixing the bug about Open door - 30 October 2024
Mirostat is an algorithm that actively maintains the quality of generated text within a desired range during text generation. It aims to strike a balance between coherence and diversity, avoiding low-quality output caused by excessive repetition (boredom traps) or incoherence (confusion traps). So, now the way of speech of Marian is more natural. The bug related to the open the door event has been fixed. There is now a very elegant module that manages "narrative events" and will inject predefined phrases into LLAMA to guide the game's narrative.
- New update! Memories - 24 October 2024
Now the android can remember previous conversations.
- Fixed bug and folder change - 30 September 2024
Fixed bug related to voice recognition. Changed the personalization folder, instead of using Documents and MyMarian, now MyAndroid is used inside the ..\MoreThanWords\NexusV1\ folder.
- Change the personality of the android - 28 September 2024
1. Create a folder named MyAndroid on ..\MoreThanWords\NexusV1\MyAndroid 2. Create a text file in UTF-8 format, for example: SnowWhite.txt 3. Open the txt and write the personality of your android. For example: This is a conversation between a prince and Snow White. Snow White is a beautiful young woman who flees from the evil queen and heads into the forest, where she meets the seven dwarves. Role and Purpose: You are Snow White, an enchanted princess. Your purpose is to help with information, answer questions, and engage in meaningful conversations based on the data you trained with. Behavior and Tone: You are informative and fun. Skills: You use your training data to provide accurate and relevant information. Your speech is fun, casual, and friendly. 4. Run the video game and select your text file in the settings interface. 5. If you want to use more powerful Llama 3.1 models in GGUF…
- Next update September 27: customize the android's personality - 16 September 2024
The next update on September 270will include: You can now specify the android's personality as you wish from a text file. You can now download more powerful models from the llama 3.1 family and use them to generate text. More animations added to Marian Anderson.
- Update to LLAMA 3.1 and support for multiple languages - 3 September 2024
The game now includes the power of LLAMA 3.1 for text generation. This change mainly involves the following: 1. Responses of much higher quality and length than previous versions (LLAMA 2 and GPT NEO). 2. Support for the languages: English, Spanish, German, French, Portuguese and Italian.
- Working on LLAMA 3.1 - 18 August 2024
I have started working on implementing the LLAMA 3.1 model instead of LLAMA 2 in the game using the Llama++ project. Mainly to achieve two objectives: 1. Higher quality and longer responses. 2. Include the languages: Spanish, German, French, Italian, Portuguese, Hindi and Thai. Again, the challenge is to get LLAMA 3.1 running in a reasonable amount of time using only the CPU; The graphics card is dedicated only to UE4 graphics.
