New Block Diffusion Plugin Creates AI-Generated Images from Text Prompts

AI image generation has quickly become more accessible over the past few months and is making its way into different industries that require asset creation. Although it’s still somewhat limited, AI image generators can create breathtaking photorealistic images from complex prompts. AI-assisted artists have even begun selling their works on stock photography sites.

DALL-E 2 was the first to make a big splash on social media. Its closed platform and invite-only beta has kept many potential creators on the sidelines, but open source alternatives like Stable Diffusion are taking off. The machine learning, text-to-image model is improving thanks to its growing ecosystem of contributors and extenders.

Kevin Batdorf, an Application Developer at Extendify, has created a free plugin called Block Diffusion that uses Stable Diffusion to generate images inside the block editor based on text prompts.

Block Diffusion uses the public Replicate API to generate images in the cloud. This requires users have an account at replicate.com. It will generate some images for free but thereafter will be $0.01 USD per image using Replicate’s Stable Diffusion model. Images generate in six seconds.

The model was trained on LAION-5B, an open AI training dataset containing over five billion image-text pairs. It has some limitations in that it cannot achieve perfect photorealism or render legible text. It also may have difficulty rendering faces and people.

For many WordPress content use cases, the Stable Diffusion model should be adequate for illustrating posts and creating unique featured images. Batdorf plans to add the following features in future releases:

  • Use an existing image as starting point
  • Add an image mask to only generate specific areas
  • Help with prompt ideas and inspiration
  • Run multiple image predictions at once
  • Have more control over the image generation steps
  • View your prediction history
  • Model selection from collections

The Block Diffusion plugin is a personal side project for Batdorf and he doesn’t plan to turn it into a commercial plugin.

“There are no plans to lock any features behind a pro or premium plugin,” he said. “All features will be added to this version on wordpress.org. Over time I plan to add all features supported by Replicate, as well as additional features for the WordPress community specifically.”

In the meantime, Batdorf said users who find the plugin useful can support its continued development by sponsoring the project on GitHub through monthly or one-time contributions .

5

5 responses to “New Block Diffusion Plugin Creates AI-Generated Images from Text Prompts”

  1. I have come across this tech only recently. My friend is using AI to generate NPC images for RPG sessions.

    I can see this working in a number of industries though. Police could use this tech to build criminal profiles on the basis of witness reports. Historians could generate images of historical figures based on old writings. Exciting!

  2. This is awesome, and the $0.01 USD per image is actually a great price, compared to alternatives or self-hosted Python. Can’t wait to see more updates and, eventually, the entire stack going free/open-source.

  3. This is great news. I am on the A.I. image train since a couple of weeks now by using midjourney. It definitely helps keeping costs low compared to buying stock images. However, some of the big stock image sites have closed submissions for computer generated graphics due to copyright issues. We should wait to see how this works out.

  4. The only concern I have is that many of these systems use collections of images where the original artist may not have given permission for the artwork to be used that way. It really falls into a copyright gray zone and ethically questionable. Also who owns the image that is created? The original artists whose artwork was used to generate it? The service that runs the AI that generated the artwork? The user who provided the prompts to generate the artwork? Everyone involved?

    • Hmm… the commercialization of the trained weights, which were trained using a wide array of images from various inline gallery sites… is kinda wild west right now?

      As for current ownership from services, that depends on the tos/eula of that services and… of which ever country one lives and operates in.

      Are the trained weights a derived work? If so, are catalogs and databases of artwork likewise considered derived works? What level of association of features, imagery, textual information… crosses the line from inspiration to derived work?

      I have a custom SD setup at home. I am exploring imagery that while it might be in the style of xyz… isn’t a subject matter that xyz has touched or has any plans to touch. No imagery like it existed before. It is associative artwork surfaced from the impression of millions of images from the internet.

      I think short of using the tech to copy someone’s style and claiming it as your own… or representing that work as being that of the original artists’…. There is a very large grey area that will have repercussions in other areas of copyright and internet usage. (Indices, databases, security hashes, etc.)

Leave a Reply to Musti Cancel reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Newsletter

Subscribe Via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

%d bloggers like this: