Generative AI Models for Creatives

Artificial Intelligence (AI) is finding its way into all areas of society. The professional fields of the creative industries (product designers and fashion designers, composers and advertising copywriters, photographers and video producers, architects and graphic and UX designers) are not excluded from this transformational process. On the contrary, Creative AI is one of the hotspots of this development: machines that can create texts, visuals, music compositions, fashion designs and architectural designs, are not only changing the tasks and the self-image of creative people, but also have a significant influence naturally on the output - the drafts, designs and images.

This is a highly relevant social issue for several reasons. Experts suggest that in the near future, large areas of our media will be characterized by so-called synthetic content. This does not refer to the modifications of images in advertising and on social media that are commonplace today. But texts written by machines and used as product descriptions or sports reports, architectural floor plans and elevations created by algorithms, images of objects, landscapes, people, that never existed, but are deceptively created by Creative AI in a matter of moments. With this power of AI-Enabled tools, even larger groups of people gain access to creative tools that enable them to create their own creative work and to distribute it on digital platforms and, in some cases, to market it. The rapidly growing Creator Economy, in which not trained creatives, but rather amateurs produce attention-grabbing content is a first social result of this Creative AI - development is in its infancy.

However, these developments are also giving creators access to tools that were previously only available to AI professionals and techies. No-code is the keyword here, where functions are preconfigured so that, for example, designers and creatives can control them without having to program - just as a designer doesn't need to be a programmer to get results with Photoshop. However it will probably be some time before Visual AI tools are formatted in the same way as image editing programs for the PC are today - after many years of development.

This is precisely where the CogniWerk research and transfer project comes in. The focus is on so-called AI models, which are available in various file formats (pkl, mlmodel, pb, onnx, h5, pmml, csv, json). Such AI models can, for example, create images or videos, or manipulate them. They are mostly created and pre-trained by interested software engineers and companies and made available in different ways, often not very structured, on the net.
The provision is often explicitly non-commercial. Such a pre-trained AI model can, for example, increase very low image resolution of an existing image without loss. Another such a model can create finished building views from line drawings. There are countless of these CreativeAI models and the supply is growing daily.

Many of the models are published as papers on university websites, while other models are featured and discussed by users on Reddit forums. Models end up on Github.com and comparable repositories, etc. Due to the rapidity of development, it is difficult for the non-specialist to keep track. Furthermore, it is hard to understand what exactly the capabilities of an AI model are and which requirements (e.g. on installation, technical integration) have to be considered. As a rule the models have extensive installation instructions and considerable prior knowledge is required to achieve an output. As valuable as this wealth of offerings is in principle, it is not very accessible to those who could benefit most: the creatives. Most models are made available by techies. The interest behind them is often a technical one and not in the application. Accordingly "insufficient" is the preparation of the models' capabilities (e.g. example photos) and a manageable technical documentation. In most cases, the models must not only be downloaded, but then installed on a powerful cloud computer and operated there.

The concept of CogniWerk promises a structured capture of many models, so that users, creatives from different disciplines, can be provided with specific models and who are suggested to them. The functionality and possibilities of the models will be illustrated by a preview function. A large number of these models can then also be used

immediately by creatives - without coding - for their own purposes.
CogniWerk lists Creative AI models and provides suitable recommendations for the creative's use case.

Our currently listed models include: Dall-E 2, sounddraw.io, neural.love, Centepede Diffusion, LookingGlas1.5, Superresolution with GPEN, Structured Dreaming - Styledream Faces, Mixed Reality with GANgealing, Deep Painterly Harmonization, artflow.ai, CLIPasso, Prosepainter, NeuralStyleTransfer, LaMa Inpainting, Superresolution with GFPGAN, ruDALL-E, minDALL-E, SculptGL, Multi-Perceptor CLIP Guided Diffusion, 360 Diffusion, Guided Diffusion, DEEP-L, ShapeMatchingGAN, Liquid Warping GAN, PIFuHD, Face Translation, First Order Model, DeOldify, Scroobly, Lobe, Teachable Machine, Face Parsing, YOLOv3, U2-Net, U-Net, Speech Command Recognizer, Toxicity Classifier, Question and Answer, PoseNet, BlazePose, MoveNet, Handpose, Hand Pose Detection Face Landmarks Detection, Face Detection, Depth Estimation, BodyPix, Blazeface, coco-ssd, MobileNet, BERT, GPT-2, The Infinite Drum Machine, Aiva, Melobytes AI Song Melobytes AI Music, Text Detection in Image, Image to Sound,, Image to Song, AI Image Recognition, AI Celebrities Recognition, AI Become a singer, Skindeep, Ebsynth, SuperResolution with ESRGAN, SuperShapesGenerator, SuperResolution with LatentDiffusion, AttnGAN, Dall-E, SketchRNN, CycleGAN, Sketchmachine, Artbreeder, Synthesia, Monstermash, Pix2Pix, VQGAN + CLIP, BigGAN, GPT-NeoX, GPT-J, GPT-3, Prank Your Zoom-Call, NoseNet mit PoseNet, Test Docker, Image Classifier, Mixing Things, Dancing 3D-Puppet, GauGAN2