Naresh Kumar Devulapally
Talk on T2I watermarking
March 28 2025
Your Text Encoder Can Be An Object-Level Watermarking Controller
Naresh Kumar Devulapally
Invited talk: Cognitive Vision Group UMBC
Naresh Kumar Devulapally
Recap: T2I Latent Diffusion Models
March 28 2025
Talk on T2I watermarking
Image credits: [1]
Naresh Kumar Devulapally
Personalization with Textual Inversion
March 28 2025
Talk on T2I watermarking
v^* = \arg \min_v \mathbb{E}_{z \sim p(z), y, \epsilon \sim \mathcal{N}(0, 1)} \left[ \| \epsilon - \epsilon_{\theta}(z, t, c_{\theta}(y)) \|^2 \right]
Image credits: [2]
Naresh Kumar Devulapally
Watermarking Pipeline
March 28 2025
Talk on T2I watermarking
\text{Watermark Embedder } (E_{\theta})
\text{Message } (m)
\text{Watermark Detector } (D_{w})
\text{Predicted } (m')
We want to match
these by training \( E_\theta \)
Input Image \( I_0 \)
Watermarked Image \( I_w \)
Naresh Kumar Devulapally
Watermarking Pipeline
March 28 2025
Talk on T2I watermarking
\text{Watermark Embedder } (E_{\theta})
\text{Message } (m)
\text{Predicted } (m')
We want to match
these by training \( E_\theta \)
Blind Watermarking: No additional metadata other than a watermarked image is required for detection.
Input Image \( I_0 \)
Watermarked Image \( I_w \)
\text{Watermark Detector } (D_{w})
Naresh Kumar Devulapally
Robust, Invisible watermarking constraints
March 28 2025
Talk on T2I watermarking
Payload
Robustness
Invisibility
Naresh Kumar Devulapally
Robust, Invisible watermarking constraints
March 28 2025
Talk on T2I watermarking
Payload
Robustness
Invisibility
- Watermarking within T2I generation.
- Object Level watermark localization.
- Towards training-free methods.
- Natural control for watermarking.
Additional constraints while using
Generative models for watermarking.
Naresh Kumar Devulapally
Robust, Invisible watermarking constraints
March 28 2025
Talk on T2I watermarking
- Watermarking within T2I generation.
- Object Level watermark localization.
- Towards training-free methods.
- Natural control for watermarking.
\textcolor{blue}{\text{Integrate Watermarking into the generation pipeline}}
\textcolor{blue}{\text{Watermark localization during image generation}}
\textcolor{blue}{\text{Very low parameter requirement}}
\textcolor{blue}{\text{User facing components during generation}}
Naresh Kumar Devulapally
Our Contributions
March 28 2025
Talk on T2I watermarking
\textcolor{green}{\text{Integrate Watermarking into the generation pipeline}}
\textcolor{green}{\text{Watermark localization during image generation}}
\textcolor{green}{\text{Very low parameter requirement}}
\textcolor{green}{\text{User facing components during generation}}
\textcolor{blue}{\text{Text controlled}}
\textcolor{blue}{\text{768 params}}
Naresh Kumar Devulapally
Our Method
March 28 2025
Talk on T2I watermarking
L_w = \text{BCE}(D_w(\text{Dec}(z'_0, w))), m)
\mathcal{D}_w - \textcolor{blue}{\text{Message Detector}}
z'_0 - \textcolor{blue}{\text{Watermarked Latent}}
m - \textcolor{blue}{\text{Watermark Key}}
\text{(Binary string) i.e., } m \in \{0, 1 \}^k
Naresh Kumar Devulapally
Our Method
March 28 2025
Talk on T2I watermarking
L_z = \min_{\mathcal{W}_*} \mathbb{E}_t \left[ \| z_t^* - z_t(\mathcal{W}_*) \|^2 \right]
z_t - \textcolor{blue}{\text{Latent at }} t
\mathcal{W}_* - \textcolor{blue}{\text{Watermark token}}
Naresh Kumar Devulapally
Our Method
March 28 2025
Talk on T2I watermarking
Inference
Naresh Kumar Devulapally
Datasets and Metrics
March 28 2025
Talk on T2I watermarking
Datasets:
MS COCO, WikiArt
Image similarity:
PSNR, SSIM, FID
Robustness Metrics:
Basic and Adversarial Attacks
Naresh Kumar Devulapally
Cross Attention Maps
March 28 2025
Talk on T2I watermarking
Naresh Kumar Devulapally
Cross Attention Maps
March 28 2025
Talk on T2I watermarking
Naresh Kumar Devulapally
Cross Attention Maps
March 28 2025
Talk on T2I watermarking
Multiple Object Watermarking
Naresh Kumar Devulapally
Plug and Play with SD pipelines
March 28 2025
Talk on T2I watermarking
Naresh Kumar Devulapally
Qualitative Results
March 28 2025
Talk on T2I watermarking
Naresh Kumar Devulapally
Experimental results
March 28 2025
Talk on T2I watermarking
Naresh Kumar Devulapally
Experimental results
March 28 2025
Talk on T2I watermarking
Naresh Kumar Devulapally
References:
March 28 2025
Talk on T2I watermarking
- High-Resolution Image Synthesis with Latent Diffusion Models.
- An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion.
- The Stable Signature: Rooting Watermarks in Latent Diffusion Models.
- AquaLoRA: Toward White-box Protection for Customized
- Stable Diffusion Models via Watermark LoRA.
Invited talk on T2I Watermarking: UMBC
By Naresh Kumar Devulapally
