Jump to content

  •  

Picsart Introduces Open-Source AI Art Tool With Firefly-Like Abilities


Community

Recommended Posts

Could contain:

Video screenshots via Picsart

Picsart’s AI Research team has introduced HD Painter, an open-source artificial intelligence project aimed at enhancing the precision and quality of image inpainting with high-resolution, realistic results through text-to-image diffusion models.

Image inpainting, a computer vision technique, helps reconstruct missing or unwanted parts of images in a seamless manner. Whether it’s a spot, a tear, or an object in a photograph, inpainting intelligently fills these gaps using the surrounding image data. 

However, Picsart asserts that existing methods by software like Stable Diffusion, Google Imagen, and DALL-E 2 often lack precision and might not always align with the user’s intentions. For example, if you tell the program to add a “cat” to an empty part of the photo, it might somehow miss that—engaging in something called prompt neglect—and just fill in with more background or other objects from the image.

HD Painter steps up to this challenge by being more sensitive to and executing text instructions or prompts, thus completing the missing parts of an image accurately and in high resolution.

Those who have experimented with Generative Fill in Photoshop, powered by Adobe’s generative AI model Firefly, would be familiar with this concept. You first select or draw over an area of the image you want to edit or improve. This could be removing an unwanted object, filling in missing parts, or changing something existing in the shot. Then, enter a text prompt to describe what you want in the selected area. The AI analyzes the surrounding area of the selected part to understand the texture, color, and patterns of the nearby image data—crucial for making the new part blend in naturally with the rest.

Could contain:

Video screenshots via Picsart

Central to HD Painter are two innovative technologies: Prompt-Aware Introverted Attention (PAIntA) and Reweighting Attention Score Guidance (RASG). PAIntA improves the model’s attention to user’s text prompts, ensuring that the reconstructed area closely matches the instructions. Meanwhile, RASG adjusts the inpainting process to align precisely with the text prompts, leading to higher quality and accurate results.

A notable advantage of HD Painter is its capability to achieve up to 2K resolution inpainting, offering a substantial improvement in clarity and detail over other methods. This feature ensures that the final images are not only complete but also sharp and refined.

Interestingly, HD Painter employs a “training-free” methodology—meaning it only utilizes pre-trained models and adaptable components, with no additional data needed. This minimizes the need for additional training and computational costs and time, while enabling the tool to be applied to any existing diffusion-based inpainting model.

Find more about the feature here.

 

 

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.





×
×
  • Create New...
The Creative Network

DesignTAXI

The Creative Finder

The Bazaar

Trendingger (BETA)

Community Resources

Become a member

  • Sign up for free
  • Pro/Business Accounts
  • Log into your account

    Forum Rules & Guidelines

    Terms of Use

    DMCA Copyright Notice

    Privacy Policy

    Cookies

    Contact Us

    Advertise with us

  • Express self-serve ads
  • Other advertising inquiries
  • Popular Categories

  • Artificial Intelligence (AI)
  • Innovation
  • Accessibility
  • Creative Ad Ideas
  • Climate Change & Sustainability
  • Copyright
  • Humor
  • Inclusivity
  • Travel
  • WTF
  • Creative Disciplines

  • 3D
  • AR / VR
  • Architecture
  • Art
  • Automotive
  • Branding
  • Character Design
  • Comics
  • Fashion Design
  • Furniture Design
  • Graphic Design
  • Illustration
  • Industrial Design
  • Interior Design
  • Logo Design
  • Packaging Design
  • Product Design
  • Street Art
  • Typography
  • UI/UX
  • Video Games