Support image/video content analysis (multimodal input)

[l2dnjsrud]

Description

Currently PhantomCrowd only analyzes text content. Many marketing campaigns are image/video-based. Adding multimodal input would dramatically improve accuracy.

From the roadmap

This is listed in README roadmap as a planned feature.

Proposed approach

• Accept image uploads via the campaign creation form
• Use a vision-capable LLM (e.g., gemma4, llava) to describe the image
• Feed the description into the existing pipeline as additional context
• Display uploaded images in the campaign detail view

Technical notes

• Backend: add image upload endpoint, store in data/uploads/
• LLM: use Ollama vision model to generate description
• Frontend: add image preview in campaign form and detail view

Difficulty

Intermediate. Requires backend + frontend + Ollama vision model integration.