Kiel-2-Omni

Kiel-2-Omni is a fine-tuned multi-modal vision-language model based on the Janus architecture. It unifies three core capabilities within a single autoregressive framework:

1. Text Generation: Standard conversational reasoning and open-ended text answers.
2. Image Understanding (Image-to-Text): Visual question answering, scene description, and multimodal analysis.
3. Text-to-Image Generation: Generating high-fidelity visual assets and creative artwork from descriptive text prompts.

Model Details

• Model Type: Multi-Modal Causal Language Model (MultiModalityCausalLM)
• Base Model: deepseek-ai/Janus-Pro-1B
• Architecture Highlights: Decoupled visual encoder pathways for understanding combined with a vector-quantized token decoding head for native image synthesis.
• Weights Format: Single unified .safetensors file (model.safetensors) designed for optimized cloud deployment and fast loading.

Quick Start: Loading the Model

Load your fine-tuned model and processor directly from Hugging Face using standard transformers:

import torch
from transformers import AutoProcessor, AutoModel

repo_id = "kiel2/Kiel-2-Omni"

processor = AutoProcessor.from_pretrained(repo_id, trust_remote_code=True)
model = AutoModel.from_pretrained(
    repo_id, 
    trust_remote_code=True, 
    torch_dtype=torch.bfloat16, 
    device_map="auto"
)
print("Kiel-2-Omni loaded successfully!")