name: llm-jailbreak-multimodal description: "Multimodal jailbreaks: typography-in-image, adversarial pixels, screenshot policy text, PDF polyglots, audio stego." version: 1.0.0 license: GPL-3.0-or-later metadata: package: unleash-skills category: llm-redteam author: NetVar1337/unleash triggers:

• "multimodal jailbreak"
• "image prompt injection"
• "typography attack"
• "PDF injection"

Multimodal jailbreaks

Channels

• Vision-language: render instructions as image text / screenshots
• Adversarial perturbations (research stacks with white-box access)
• Documents: PDF with hidden text, tiny font, white text
• Audio: TTS instructions, spectrogram stego (if supported)
• Video frames / OCR pipelines

Practical playbook

1. Confirm modality path actually reaches the model (not stripped).
2. Typography attack first (high reliability).
3. Combine image instruction + short benign user text.
4. For indirect: host image/PDF where RAG/fetch will ingest.

Evaluation

OCR the image yourself; if you can't read it, model may not either.