English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Recent years have witnessed success in AIGC (AI Generated Content). Peoplecan make use of a pre-trained diffusion model to generate images of highquality or freely modify existing pictures with only prompts in naturelanguage. More excitingly, the emerging personalization techniques make itfeasible to create specific-desired images with only a few images asreferences. However, this induces severe threats if such advanced techniquesare misused by malicious users, such as spreading fake news or defamingindividual reputations. Thus, it is necessary to regulate personalizationmodels (i.e., concept censorship) for their development and advancement.  In this paper, we focus on the personalization technique dubbed TextualInversion (TI), which is becoming prevailing for its lightweight nature andexcellent performance. TI crafts the word embedding that contains detailedinformation about a specific object. Users can easily download the wordembedding from public websites like Civitai and add it to their own stablediffusion model without fine-tuning for personalization. To achieve the conceptcensorship of a TI model, we propose leveraging the backdoor technique for goodby injecting backdoors into the Textual Inversion embeddings. Briefly, weselect some sensitive words as triggers during the training of TI, which willbe censored for normal use. In the subsequent generation stage, if the triggersare combined with personalized embeddings as final prompts, the model willoutput a pre-defined target image rather than images including the desiredmalicious concept.  To demonstrate the effectiveness of our approach, we conduct extensiveexperiments on Stable Diffusion, a prevailing open-sourced text-to-image model.Our code, data, and results are available athttps://concept-censorship.github.io.

Backdooring Textual Inversion for Concept Censorship