PIXART-Delta: Instant Text-to-Image Generation Model.

Written By: kinza.sabir
Last Updated On: January 19, 2024

Unlock the magic of PIXART-Delta (PIXART– δ) – where every stroke is a burst of brilliance! The model has the ability to create diverse range of high quality images with in 0.5 seconds which make this model standout compared to other text-to-image generation models. Researches from Huawei Noah’s Ark Lab, Dalian University of Technology, IIIS, Tsinghua University and The University of Hong Kong presented this instant masterpiece.

The model PIXART-δ takes text as an input and generates output in the form of an image. The user can input different image styles such as cinematic, photographic, 3D model, digital art and many more. The user can also customise the size of the output image which is an amazing feature for the users. With all these advanced features the generated output is of high quality and high resolution.

This image is generated using demo presented by PIXART-δ on HuggingFace. The text prompt for the above mentioned image is “professional portrait photo of an anthropomorphic cat wearing fancy gentleman hat and jacket walking in autumn forest.“

PIXART– δ is an advanced version of PIXART-α which is extremely efficient in creating high-quality images with the resolution of 1024 pixels. It has the ability to create images in 2-4 steps which is really quick. Whereas, PIXART-α was providing the same output but it was taking a lot more steps to generate the required output.

The exceptional feature of PIXART-δ is that it has the ability to generate image in just 0.5 seconds, which is super fast. It is 7 times quicker than PIXART-α in terms of memory, time and computational cost.

The above added clip is combination of different images generated by same text prompt using demo at HuggingFace. The model PIXART– δ is extremely creative that it provide the user with diverse range of images. The given input was “Kids playing in the garden, rainy day, swings, sun, rainbow” and the generated output are from different genres such as Cinematic, Photographic, Anime, Digital Art, Manga and 3D Model. This model provide the user with the facility of download to get the images without compromising on the quality of the images.

Technicalities of PIXART-Delta

Latent Consistency Model (LCM) and ControlNet was incorporated into the advanced PIXART-α model which eventually forms PIXART– δ. The incorporation of Latent Consistency Model (LCM) make the PIXART-δ text-to-image generation process better and faster and PIXART-δ is able to generate high-quality and high resolution images in just 0.5 seconds.

ControlNet has good control over text-to-image generation which helps to make the right adjustments and follows the instructions to generate the desired output. It has good control over the complete layout of the generated image.

The text prompt for the above added image “stars, water, brilliantly, gorgeous large scale scene, a little girl, in the style of dreamy realism, light gold and amber, blue and pink, brilliantly illuminated in the background.” with the genre of “photographic” is generated by demo at HuggingFace.

The text prompt for the above added image is also “stars, water, brilliantly, gorgeous large scale scene, a little girl, in the style of dreamy realism, light gold and amber, blue and pink, brilliantly illuminated in the background.” with the genre of “Anime” is generated by the demo of PIXART– δ at HuggingFace.

From both of the images above, it is very clear that the model generates different images even when the same text input is provided to the model. This shows that PIXART– δ is extremely creative and diverse.

Wrap Up!

PIXART-δ is an efficient text-to-image model that can create really nice and detailed pictures that are 1024 pixels in size in just 1 second. The key advantages lie in the model’s speed, high image quality, and fine-grained control, opening up new possibilities in various domains such as virtual reality, content creation, interactive user interface etc.