ComfyUI - custom AI Mosaic workflow
Budget: $750 – $1,500 USD
I am looking for an advanced ComfyUI specialist that can create a custom CUI workflow that would transform images based on 3 inputs:
- modelling image (explained below)
- person image
- prompt
This workflow will be a part of a new event attraction: AI Photo Mosaic. A regular photo mosaic is a wall of hundreds of photos creating a final mosaic image. The final image is created from the individual tiles placed at specific tiles because each tile has a semi-transparent overlay over each of the photo booth source images that is the main image's part in this specific tile.
Examples and process description of a regular mosaic can be found on my site: https://www.imprero.com/en/mosaix-photo-mosaic-wall/
We are planning a new version of the mosaic - an AI mosaic, which is different from the regular mosaic in a way, that each tile does not have any semi-transparent overlay put on the photo booth source images, but each source image is sent to an AI workflow that transforms this image into a mosaic tile in a way, that the colors and shapes of the AI created image "fit" into the general mosaic image as its "pixel". The best example of a final version of such mosaic is this: https://amp.livemosaics.com/universal-ai/. There you can zoom in and see how each image (and its colors) are part of the bigger picture.
The new custom ComfyUI workflow would therefore need to accept 3 inputs mentioned above, where the "modeling" image is the specific tile of the main target image which the workflow needs to use as base for the output image colors and shapes. The example mosaic creates the final ai mosaic where individual images have only specific colors applied - which is sufficient for larger mosaics that do now have small details. However we would like that the requested workflow applies also shaping the transformed images, so that not only colors but image shapes reflect the modelling tile - this would allow to create a more sharp target image with smaller details with fewer tiles.
Keep in mind that the colorization of the ai transformed image needs to be based ONLY on the modelling image data, not on the prompt (as the prompt will have only style description, eg. "photo of a woman, a superhero in a futuristic suite").
Key requirements:
- Real-time image processing (expected total image processing time max. 15s/image)
- Input by json file consisting of base64 converted images (modelling image and person image) and styling prompt
- Output in json format with the processed image in base64 format
The workflow will eventually run on a Runpod serverless endpoint (RTX4090 or stronger).
We will consider only proposals from freelancers that have actually read this description. In order to prove this, start your offer/reply with the word 'read' in capital letters. Additionally, we will most probably not consider offers that are submitted within the first 5 minutes of the project publication - as these will be treaded as auto generated, thus unreliable!
The freelancer is expected to provide a testing environment available over the Internet. The project will be divided into 1 milestone: the workflow will be converted to a Runpod serverless endpoint and quality and processing times verified (RTX4090 or stronger, times verified on an active worker).
- modelling image (explained below)
- person image
- prompt
This workflow will be a part of a new event attraction: AI Photo Mosaic. A regular photo mosaic is a wall of hundreds of photos creating a final mosaic image. The final image is created from the individual tiles placed at specific tiles because each tile has a semi-transparent overlay over each of the photo booth source images that is the main image's part in this specific tile.
Examples and process description of a regular mosaic can be found on my site: https://www.imprero.com/en/mosaix-photo-mosaic-wall/
We are planning a new version of the mosaic - an AI mosaic, which is different from the regular mosaic in a way, that each tile does not have any semi-transparent overlay put on the photo booth source images, but each source image is sent to an AI workflow that transforms this image into a mosaic tile in a way, that the colors and shapes of the AI created image "fit" into the general mosaic image as its "pixel". The best example of a final version of such mosaic is this: https://amp.livemosaics.com/universal-ai/. There you can zoom in and see how each image (and its colors) are part of the bigger picture.
The new custom ComfyUI workflow would therefore need to accept 3 inputs mentioned above, where the "modeling" image is the specific tile of the main target image which the workflow needs to use as base for the output image colors and shapes. The example mosaic creates the final ai mosaic where individual images have only specific colors applied - which is sufficient for larger mosaics that do now have small details. However we would like that the requested workflow applies also shaping the transformed images, so that not only colors but image shapes reflect the modelling tile - this would allow to create a more sharp target image with smaller details with fewer tiles.
Keep in mind that the colorization of the ai transformed image needs to be based ONLY on the modelling image data, not on the prompt (as the prompt will have only style description, eg. "photo of a woman, a superhero in a futuristic suite").
Key requirements:
- Real-time image processing (expected total image processing time max. 15s/image)
- Input by json file consisting of base64 converted images (modelling image and person image) and styling prompt
- Output in json format with the processed image in base64 format
The workflow will eventually run on a Runpod serverless endpoint (RTX4090 or stronger).
We will consider only proposals from freelancers that have actually read this description. In order to prove this, start your offer/reply with the word 'read' in capital letters. Additionally, we will most probably not consider offers that are submitted within the first 5 minutes of the project publication - as these will be treaded as auto generated, thus unreliable!
The freelancer is expected to provide a testing environment available over the Internet. The project will be divided into 1 milestone: the workflow will be converted to a Runpod serverless endpoint and quality and processing times verified (RTX4090 or stronger, times verified on an active worker).