New tools for ultimate character consistency — from heatherbcooper.substack.com by Heather B. Cooper
Plus simple & effective video prompt tips

We have some new tools for character, objects, and scene consistency with Runway and Midjourney.

Multimodal AI = multi-danger — from theneurondaily.com by Grant Harvey

According to a new report from Enkrypt AI, multimodal models have opened the door to sneakier attacks (like Ocean’s Eleven, but with fewer suits and more prompt injections).

Naturally, Enkrypt decided to run a few experiments… and things escalated quickly.

They tested two of Mistral’s newest models—Pixtral-Large and Pixtral-12B, built to handle words and visuals.

What they found? Yikes:

    • The models are 40x more likely to generate dangerous chemical / biological / nuclear info.
    • And 60x more likely to produce child sexual exploitation material compared to top models like OpenAI’s GPT-4o or Anthropic’s Claude 3.7 Sonnet.

Rise of AI-generated deepfake videos spreads misinformation — from iblnews.org