Muse 1.0 launched last year on the iPad as a spatial and visual space for deep thinking. Computers are more often involved in the later stages of the creative process: production work like writing a book, designing an interface, or creating a CAD diagram. This is somewhat unprecedented: typically thinking work is done with analog tools like a sketchbook, a whiteboard, index cards, or post-its. Muse is a digital workspace for thinking. Read on for the full vision, and how you can help. The txt2image prompt I used was: ‘picture of nigel adams spcman as neo from the movie matrix, green digital background, badass’ lol.Muse 2.0 will include a native Mac app and local-first sync between iPad and Mac, launching in spring of 2022. The pictures below were generated in DiffusionBee (with the custom model option). This may be inside a subfolder and will be approximately 2 GB or more. After running the cellsĪt the end of the process navigate to the OUTPUT_DIR in your gdrive and find your checkpoint model. The training step (starting !accelerate ) will take 30 mins or so. Run the rest the without changing the code/defaults. save_sample_prompt="photo of nigel adams spcman" \ Refresh the files list and drag and drop your training images as per below. After running this cell, the code will generate a folder to upload your training imagess. Obviously remember the instance prompt as that will trigger your image in the AI art generator. Once you have the model you can download it and use it in other experiments or programs like DiffusionBee to generate the AI art. This is the output directory on gdrive where the check-point model will be saved. OUTPUT_DIR stable_diffusion_weights/nigeladams HUGGINGFACE_TOKEN: xxxxx (paste your own token from hugging face after generating) Settings and Run Here are the code blocks I updated (as per the instructions on the notebook). You don’t need programming skills as such and can use the checkpoint model you create in DiffusionBee for fast results without any programming skills. Open this up….but note I’m not going to do a full tutorial as you’ll find good instructions quite easily by searching for ‘DreamBooth Colab’ on Google or YouTube. The DreamBooth Colab notebook I used is here jpg as the file format and file extension. Try to get a few varying angles of the subject. Don’t worry about removing the background or any fancy photo editing. The images must be 500 pixels x 500 pixels. Start by creating a new folder on your computer and add 10 or more input images of yourself (or the subject you want to use). You just have to create an account and read/accept the terms. The only other prerequisite is having a HuggingFace access token but this is free also. Having Google Drive (gdrive) setup is handy so you can permanently save the new check-point model generated. The whole training process takes about 30 mins. This is cheating in the context of my blog, however, training a new model is an infrequent task so why spend too long figuring it all out on the Mac when a Colab notebook already does the job. I did try to get this running locally on my Mac but ended up using Google’s Colab for this experiment. Unoriginally I thought I would experiment by feeding it images of me! Feeding it only 15 pictures I was still blown away. With DreamBooth you can fine-tune and add your own images into the AI art’s learning model.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |