Poketex is a web platform that provides users with tools to generate and interact with Pokémon characters created through AI.
It is the product of a bachelor's thesis from the Faculty of Computer Science in Iasi and an extension of a project that won at FIIPractic.
Link to the paper.
- Main Page: Showcases the most popular Pokémon and includes search functionality with pagination.
| Home |
|---|
 |
- Rating and Comments: Users can rate ⭐⭐⭐⭐⭐ and comment on each Pokémon.
- MonBuilder: AI-Based Pokémon Creation tool where users can specify a prompt for how the Pokémon should look, a negative prompt for excluding certain elements, a seed for precise and consistent generation, and the number of steps for processing intensity.
| Home |
|---|
 |
- PokePaint: A drawing interface where users can sketch their ideas. The AI interprets these sketches to generate a Pokémon image.
- Paint: Real-Time canvas with the ability to draw, Erase, Change color or the Brush size, Undo/Redo and image uploading.
- ControlNET Integration: Enhances image uploads by transforming them into pencil sketches with "control_v11p_sd15s2_lineart_anime" preprocessor.
- NSFWJS: Ensuring the content is Safe for Work testing the image for malicious or +18 content.
| Paint Components | NSFWJS (Adult Content Filtering) | Image Uploading with ControlNET Preprocessing |
|---|---|---|
 |
 |
 |
- StickPose: A stickman editor allows the user to pose figures with adjustable joints. Depth Maps and image segmentation are employed to add and layer objects, creating a scene for the AI to interpret.
- Allows for the adding or removal of any stickman from the scene.
- Users can add objects to the scene.
- Stickmen and objects can be moved forwards or backwards in the scene.
- Offers settings to adjust how much the pose affects the generated image.
- Includes a function to save and load poses into/from a database.
| StickPose examples | Image Generation | Saved Poses |
|---|---|---|
 |
 |
 |
| Technologies for Image Segmentation | DepthMaps Objects Menu | Overlapping the Generation with Objects |
|---|---|---|
 |
 |
 |
- ChatGPT Integration: Using the ChatGPT API and the prompt of the Pokemon, it generates a name and a backstory of the character.
| LLM Generated Text |
|---|
 |
- Profiles: Users have their profiles where their generated Pokémon are displayed with unique names and stories.
- Breeding: Users can combine two Pokémon, one from their collection and another from their own or another user's collection.
| User Profile | Breeding |
|---|---|
 |
 |
- Pokémon Pages: Each Pokémon has a dedicated page featuring its story, attributes, and community ratings.
| Pokemon |
|---|
 |
Frontend
React React official website
Konva-React Konva-React official website
TypeScript TypeScript official website
Backend
API Java Spring Boot
Database
PostgreSQL PostgreSQL official website
Stable Diffusion with LoRA and ControlNET
- Art generation is powered by the model trained by Justin Pinkney, with Lambda Diffuser and Stable Diffusion:
- Stable Diffusion Checkpoint: https://civitai.com/models/23900/anylora-checkpoint
- Stable Diffusion LoRA: https://civitai.com/models/5115/pokemon-lora-ken-sugimori-style-for-fakemon-and-characters
- ControlNET Preprocessor: https://huggingface.co/lllyasviel/control_v11p_sd15s2_lineart_anime
- Stable Diffusion module utilizes the Automatic1111 nowebui Rest API: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/API.
Other A.Is used
- NSFWSJS library to check the content of the uploaded images: https://github.com/infinitered/nsfwjs
- IS-NET for Image Segmentation: https://arxiv.org/abs/2108.12382
- ZoeDepth used to create depth maps for objects in the editor: https://github.com/isl-org/ZoeDepth
ChatGPT
- The Name and the Description of the pokemon is automatically generated with ChatGPT-3.5TURBO model : https://github.com/PlexPt/chatgpt-java
Security
- Authentication and user management is handled by Okta with OAuth2 security. https://developer.okta.com/
- Bytex for organizing the "Java in the world of microservices" training (Trainers : Alexandru Cretu, Cristian Danila)
- Justin Pinkney for the art generation model.
- Automatic1111 for the Stable Diffusion API
- PlexPt for the ChatGPT connexions
- Okta for providing user authentication and management.