Stable Diffusion Project
Budget: $250 – $750 USD
PLEASE DO NOT SIMPLY CUT AND PASTE REQUIREMENTS AS YOUR BACKGROUND EXPERIENCE, IF YOU INTEND TO BID YOU WILL BE REQUESTED TO PROVIDE EXAMPLES OF SPECIFIC DIRECT SUCCESSFUL EXPERIENCE FOR EACH OF THE LINE ITEMS BELOW!
ADDITIONALLY, THE LINE ITEMS MAY BE USED TO CREATE A MILESTONE DELIVERY SCHEDULE.
1) Download AI software for Image transfer usage Stable Diffusion or Equivalent must have inline painting as well
2) Load large image library for training and as source images and bulk batch overnight jobs, images will be in personal
generated image folders, there will be no utilization of online images.
3) Try to take advantage of RTX GPU for performance
4) Test and verify system
5)I have a personal Image library that now consists of over 100gig created Bing Image creator 1024 x 1024 images, none were generated from photos of real items they are 100% prompt created images and are based on combinations of art style theme material lighting etc variations. System must allow for additional batch loading from newly created image folders. Folder content could be Bing images, personal artwork, or photos , color or black and white, normalized to 1024 x 1024, jpg. png. tif. 8 bit color is ok, also will postprocess through upscaler to 4k to 8K after image is generated. Upscaler can be third party package.
6) There will be a library of my own ballpoint pen and outline drawings with the objective of not only coming up with new image variations but also to research if AI can capture and transfer my drawing style to the incoming images.
7) Post processing of images will go through software that will create 3D stereoscopic views for use with the ASUS 3D studio book when it becomes available and probably and Oculus Quest VR
8) If AI portion is executed there may be a contract to create a full turnkey system with all the above features.
9) Only persons with direct and PROVEN experience with generative and diffusion AI code and algorithms will be considered, so it can be Python, Kera's, TensorFlow etc and will not be set up for pay as you go credits, strictly for personal use as a private research platform
10) As much third party or open-source code that can be used should be used, this is more software integration not development from scratch, would prefer LoRa integration as well, (Detail, LineArt, Pose).
AUTOMATIC1111's(??)
11) Would prefer off the shelf best interface with little custom code
12) System should allow for image inputs and process against images without requiring a text prompt perhaps 4-8 images for the content layers, and 2 image inputs for the style layers, it is probably a dual system using a GAN for feature extraction on the content layer and diffusion for the style transfers.
13) Preference will be given to individuals who can integrate system with the nVidia CuDnn libraries.
14) Demonstration of working system and transition plan from developer to customer system
Skills Required
Software Architecture
Software Testing
Artificial Intelligence
The goal is to create an end to end AI 3D systems for Art generation, future contracts would be to also take the image outputs into an AI program to generate automatic 3D models to in .stl to go into a slicer program for creation of 3D resin prints, that software is being evaluated as well to determine if it can handle complex 2D to 3D image conversions. This is for private research there are no immediate plans to sell this, it is an R & D systems experiment
System should allow for image inputs and process against images without requiring a text prompt perhaps 4-8 images for the content layers, and 2 image inputs for the style layers , it is probably a dual system using a GAN for feature extraction on the content layer and diffusion for the style
ADDITIONALLY, THE LINE ITEMS MAY BE USED TO CREATE A MILESTONE DELIVERY SCHEDULE.
1) Download AI software for Image transfer usage Stable Diffusion or Equivalent must have inline painting as well
2) Load large image library for training and as source images and bulk batch overnight jobs, images will be in personal
generated image folders, there will be no utilization of online images.
3) Try to take advantage of RTX GPU for performance
4) Test and verify system
5)I have a personal Image library that now consists of over 100gig created Bing Image creator 1024 x 1024 images, none were generated from photos of real items they are 100% prompt created images and are based on combinations of art style theme material lighting etc variations. System must allow for additional batch loading from newly created image folders. Folder content could be Bing images, personal artwork, or photos , color or black and white, normalized to 1024 x 1024, jpg. png. tif. 8 bit color is ok, also will postprocess through upscaler to 4k to 8K after image is generated. Upscaler can be third party package.
6) There will be a library of my own ballpoint pen and outline drawings with the objective of not only coming up with new image variations but also to research if AI can capture and transfer my drawing style to the incoming images.
7) Post processing of images will go through software that will create 3D stereoscopic views for use with the ASUS 3D studio book when it becomes available and probably and Oculus Quest VR
8) If AI portion is executed there may be a contract to create a full turnkey system with all the above features.
9) Only persons with direct and PROVEN experience with generative and diffusion AI code and algorithms will be considered, so it can be Python, Kera's, TensorFlow etc and will not be set up for pay as you go credits, strictly for personal use as a private research platform
10) As much third party or open-source code that can be used should be used, this is more software integration not development from scratch, would prefer LoRa integration as well, (Detail, LineArt, Pose).
AUTOMATIC1111's(??)
11) Would prefer off the shelf best interface with little custom code
12) System should allow for image inputs and process against images without requiring a text prompt perhaps 4-8 images for the content layers, and 2 image inputs for the style layers, it is probably a dual system using a GAN for feature extraction on the content layer and diffusion for the style transfers.
13) Preference will be given to individuals who can integrate system with the nVidia CuDnn libraries.
14) Demonstration of working system and transition plan from developer to customer system
Skills Required
Software Architecture
Software Testing
Artificial Intelligence
The goal is to create an end to end AI 3D systems for Art generation, future contracts would be to also take the image outputs into an AI program to generate automatic 3D models to in .stl to go into a slicer program for creation of 3D resin prints, that software is being evaluated as well to determine if it can handle complex 2D to 3D image conversions. This is for private research there are no immediate plans to sell this, it is an R & D systems experiment
System should allow for image inputs and process against images without requiring a text prompt perhaps 4-8 images for the content layers, and 2 image inputs for the style layers , it is probably a dual system using a GAN for feature extraction on the content layer and diffusion for the style