Influencer: Empowering Everyday Users in Creating Promotional Posts via AI-infused Exploration and Customization

TL;DR

The paper addresses the challenge of everyday users creating engaging promotional posts by introducing Influencer, an AI-infused, canvas-based tool. It combines a multi-dimensional idea-generation pipeline with context-aware exploration and flexible fusion to help novices compose images and captions that align with brand messaging, tracked via a mind-map-like interface. The authors validate their approach through a within-subject study with design novices, showing higher creativity support, engagement, and post quality compared with a Google Search + Figma baseline, supported by expert assessments. The work demonstrates a practical path toward accessible, AI-assisted design for micro-entrepreneurs and offers insights into balancing automation with user autonomy, while outlining future improvements in dataset scale, model accuracy, and collaborative capabilities.

Abstract

Creating promotional posts on social platforms enables everyday users to disseminate their creative outcomes, engage in community exchanges, or generate additional income from micro-businesses. However, creating eye-catching posts combining both original, appealing images and articulate, effective captions can be rather challenging and time-consuming for everyday users who are mostly design novices. We propose Influen, an interactive tool to assist novice creators in crafting high-quality promotional post designs, achieving quick design ideation and unencumbered content creation through AI. Within Influencer, we contribute a multi-dimensional recommendation framework that allows users to intuitively generate new ideas through example-based image and caption recommendation. Further, Influencer implements a holistic promotional post design system that supports context-aware image and caption exploration considering brand messages and user-specified design constraints, flexible fusion of various images and captions, and a mind-map-like layout for thinking tracking and post-recording. We evaluated Influencer with 12 design enthusiasts through an in-lab user study by comparing it to a baseline combining Google Search + Figma. Quantitative and qualitative results demonstrate that \sysname{} is effective in assisting design novices to generate ideas as well as creative and diverse promotional posts with user-friendly interaction.

Figures (10)

• Figure 1: Influencer provides three basic types of interactive blocks on its mind-map canvas. Clicking Create New Text Block adds (A) an empty text block. Clicking Upload Image generates (B) an empty image block. Clicking Create Post produces (C) a blank post blockthat can fuse selected images and captions.
• Figure 2: System architecture and workflow of Influencer. From an input topic, the system first analyzes the input to return searched images (H) and the user can select a seed image from the search results. At the same time, the Caption Recommendation (A) digests the input topic and recommends captions from three dimensions (i.e., product, activity, advertisement). Next, Influencer can conduct image recommendation (B) based on the seed image in three dimensions (i.e., semantic, object, color) and suggest images for each dimension (G). It also supports context-aware exploration (C) based on a brand image or product/brand message to find materials aligning with the brand. Besides, Influencer allows users to customize the images via Regenerate or Mast Edit (E) and conceptually fuse design materials (i.e., images and captions) with an image or prompt reference (D). Together, it generates an aesthetically pleasing promotional post design with harmonious content. A user can further iteratively explore more design recommendations (A, B, G) and generate multiple design alternatives to share on social media (F).
• Figure 3: Ideation with Influencer: (A) From a text block, a panel providing three options (Generate Images, Generate Captions, Generate Post) for users to ideate captions or images and generate posts; (B) An assembly of image search results based on users' input topic; (C) From an image block, a panel providing four options for users to customize (Regenerate, Mask), conduct image recommendation (More Images), or produce post directly (Generate Post); (D) A panel providing related semantic keywords to help users explore diverse materials; (E) An image recommendation block displaying results in three dimensions (i.e., semantic, object, color); (F) A caption recommendation block displaying results in three dimensions (i.e., product, activity, and advertisement).
• Figure 4: Context-Aware exploration with Influencer: (A) Uploading a brand image and dragging to update the target image recommendation block and caption recommendation block; (B) Creating a text block with product message and dragging to update the target image recommendation block and caption recommendation block; (C) Updated image recommendation block; (D) Updated caption recommendation block.
• Figure 5: Customization with Influencer: In the Edit mode, an image (A) can be modified with "Mask" and "Regenerate" for customization (B). "Regenerate" produces a semantically similar new image (C). "Mask" allows for framing the part to customize with a prompt (D) and thus produces a new image (E). In the Fusion mode, images and/or text can be freely combined by dragging an image or text block with a prompt to the target image or caption (F, G, H, I).