By cryptoadmin 
February 26, 2025
The GIST Editors' notes
This text has been reviewed in keeping with Science X's editorial course of and insurance policies. Editors have highlighted the next attributes whereas guaranteeing the content material's credibility:
fact-checked
preprint
trusted supply
proofread
AI device lets customers relight portraits utilizing descriptive textual content prompts
An modern AI mannequin has been developed to create dynamic lighting results in portrait photographs and movies utilizing solely textual content enter. This know-how permits customers to regulate colours simply with descriptive prompts, equivalent to "heat, freshly cooked hen" and "icy blue gentle," eliminating the necessity for classy modifying instruments.
Professor Seungryul Bak and his group on the UNIST Synthetic Intelligence Graduate Faculty have launched Text2Relight, an AI-driven lighting-specific foundational mannequin that may carry out relighting of a single portrait picture pushed by a artistic textual content immediate as proven within the picture above.
This research, carried out in collaboration with Adobe, shall be showcased on the thirty ninth Annual AAAI Convention on Synthetic Intelligence (AAAI 2025) in Philadelphia, going down on the Pennsylvania Conference Heart from February 25 to March 4, 2025. The analysis has additionally been accepted by the Affiliation for the Development of Synthetic Intelligence (AAAI), a convention within the discipline. It’s accessible on the arXiv preprint server.
The brand new mannequin excels in expressing various lighting traits, equivalent to emotional ambiance, alongside shade and brightness, all by way of pure language inputs. Notably, it adjusts the colours of each the topic and the background concurrently, sustaining the integrity of the unique picture.
In distinction to present text-based picture modifying fashions that lack specialization in lighting knowledge and sometimes end in picture distortion or restricted lighting management, Text2Relight offers a extra refined resolution.
To allow the AI to be taught the correlation between artistic texts and lighting, the analysis group developed a large-scale artificial dataset. They utilized ChatGPT and text-based diffusion fashions to generate lighting knowledge, whereas additionally implementing OLAT (One-Gentle-at-A-Time) strategies and Lighting Switch strategies to discover varied lighting situations.
As well as, the group additional enhanced the mannequin's performance by coaching auxiliary datasets centered on shadow elimination and illumination positioning, thereby enhancing visible coherence and realism in lighting results.
Professor Bak commented, "Text2Relight holds vital potential in content material creation, together with lowering modifying time in picture and video manufacturing and enhancing immersion in digital and augmented actuality settings."
Extra data: Junuk Cha et al, Text2Relight: Inventive Portrait Relighting with Textual content Steering, arXiv (2024). DOI: 10.48550/arxiv.2412.13734
Journal data: arXiv Offered by Ulsan Nationwide Institute of Science and Expertise Quotation: AI device lets customers relight portraits utilizing descriptive textual content prompts (2025, February 26) retrieved 26 February 2025 from https://techxplore.com/information/2025-02-ai-tool-users-relight-portraits.html This doc is topic to copyright. Aside from any truthful dealing for the aim of personal research or analysis, no half could also be reproduced with out the written permission. The content material is offered for data functions solely.
Discover additional
A brand new era of AI-enabled instruments for accessible, personalised media experiences shares
Feedback to editors