ACCOUNT FOR:
lightning, pinapple, cactus, mushroom wine glass newspaper
STRIP SPECIFIC EXAMPLES FROM API PROMPT TO PREVENT BIAS INJECTION!!
ENS pinning/binding: https://pinme.eth.limo/
MVP Stillshot Tech Stack
| Purpose | Library / Tool | Notes |
|---|---|---|
| Image captioning | transformers, PIL, torch |
Uses HuggingFace BLIP |
| Metadata fetching / I/O | requests, json, os, re, time |
JSON streaming, URL normalization, retries |
| Text embedding | sentence-transformers |
Uses all-MiniLM-L6-v2 |
| Agentic workflow orchestration | langgraph |
Composable state-machine graph for structured flows |
| Local LLM-based field extraction | requests to localhost:11434 |
Assumes Ollama running LLaMA3 model |
TODO: Build full flow for real-time updates of new registered IP
For pixel art to description, best option:
Open AI with prompt:
Analyze the image and return a JSON object that describes the character in structured detail for use in semantic search and filtering. The character may be a person, animal, fruit, object, or hybrid. Include these keys:
- type: One of ["character", "animal", "object", "fruit", "hybrid", "unknown"]
- category: A short word describing what it resembles (e.g. “cat”, “apple”, “robot”, “person”)
- color_scheme: Describe the dominant color palette
- head: Describe the shape/head/ears if relevant
- eyes: Describe shape, style, and expression
- mouth: If visible, describe the expression
- accessories: List of visible accessories like glasses, hats, earrings
- clothing: Describe any visible clothing
- background: Describe the backdrop
- style: Art style (e.g., “8-bit pixel art”)
- mood: Vibe or expression
- notable_features: List of traits that stand out (color, shape, accessories)
Return valid JSON.