text-to-image-multimodel-analysis

Text-to-Image Generation with Multi-Model Analysis

Generate images from text descriptions using Stable Diffusion, analyze them using CLIP and BLIP, and perform instance segmentation using Meta’s Segment Anything Model 2 (SAM2).

Hardware Setup

This project is optimized for the following high-performance architecture:

• CPU: Intel(R) Xeon(R) Platinum 8559C
• RAM: 124 GB
• GPU: NVIDIA RTX PRO 6000 Blackwell Server Edition
• VRAM: 96 GB

Installation (GPU Only)

1. Navigate to the project and activate your environment:

   cd text-to-image-multimodel-analysis
   conda activate deeplearning
   

2. Install dependencies:

   pip install -r requirements.txt
   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
   

3. Download Models: Run the download script to fetch Stable Diffusion, SAM2, CLIP, and BLIP:

   python download_models.py
   

Running the Application

Start the FastAPI server:

cd src
python -m uvicorn main:app --host 127.0.0.1 --port 8000

Browser UI

Open http://127.0.0.1:8000/ in your browser for a fully interactive Multi-Model Studio UI. You can generate images, analyze uploads, and view segmentation masks directly.

UI Previews

Below are visual examples of the Multimodel Studio interface in action:

1. Image Generation (Stable Diffusion)

2. Image Analysis (CLIP & BLIP)

3. Instance Segmentation (SAM2)

4. Download Preview Asset

API Documentation

1. Generate Image (POST /generate)

Generates an image from a prompt, performs CLIP/BLIP analysis, and extracts SAM2 segmentation masks.

Example Request:

curl -X POST http://127.0.0.1:8000/generate \
  -H 'Content-Type: application/json' \
  -d '{"prompt":"a red bird flying over a coastal cliff", "image_size": 512}'

Response schema follows the structure in example_generate_response.json.

2. Analyze Image (POST /analyze)

Analyzes an existing image using CLIP and BLIP, and runs SAM2 segmentation.

Example Request:

curl -X POST http://127.0.0.1:8000/analyze \
  -H 'Content-Type: application/json' \
  -d '{"image_base64":"iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==", "prompt":"coastal, cliff"}'

Response schema follows the structure in example_clip_response.json.

3. Segment Image (POST /segment)

Performs SAM2 segmentation on an uploaded image, returning masks and regions of interest.

Example Request:

curl -X POST http://127.0.0.1:8000/segment \
  -H 'Content-Type: application/json' \
  -d '{"image_base64":"iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg=="}'

Response schema follows the structure in example_sam2_response.json.

Note: The base64 string in the examples above represents a minimal 1x1 black pixel PNG image, which you can use to quickly test the endpoints.

Model Configurations

• Generation: Uses runwayml/stable-diffusion-v1-5 configured for half-precision (float16) to fit in GPU VRAM and uses CUDA.
• Analysis: Uses openai/clip-vit-base-patch32 for zero-shot concept classification, alongside Salesforce/blip-image-captioning-base for natural language summarization.
• Segmentation: Uses sam2.1_hiera_base_plus (Meta SAM2) for instance segmentation.