TruthArchive.ai - Tweets Saved By @skirano

Posted - August 31, 2023 at 1:03 AM

Saved - September 4, 2023 at 9:30 PM

reSee.it AI Summary

Learn how to finetune SDXL for any desired style using replicatehq. Gather 30-50 high-quality images representing the style you want. Zip and upload them. Create an account on replicatehq, name your model, and select the Image Model option. Install necessary tools in the terminal and set up API keys. Start training with your model's name, image link, and a unique style keyword. Experiment with loralr values for training speed. Monitor progress at replicatecomtrainings, which usually takes 10 mins and costs 90 cents. Follow for more AI insights.

@skirano - Pietro Schirano

Ever wondered how to fine-tune SDXL for any style you desire? I've got the simplest guide for you using @replicatehq. Let's dive in!

@skirano - Pietro Schirano

Why fine-tune? Fine-tuning lets you specialize a model to specific styles, people, or objects, making it perfect for generating high-quality, consistent images. SDXL's capabilities are game-changing. Here an example from one of my model trained on Hiroshi Nagai art.

@skirano - Pietro Schirano

First, decide on your style and gather images. I usually use 30-50 high-quality pics. Remember, it's not about quantity but QUALITY. Include variety if you're aiming to emulate a particular style. Sometime even 10 good pics are enough.

@skirano - Pietro Schirano

Once satisfied with your image selection, zip the folder, name it, and upload it. You'll need the link later.

@skirano - Pietro Schirano

Next, create an account on @replicatehq. To initiate your model, visit: https://replicate.com/create.

@skirano - Pietro Schirano

Next, name your model. For hardware, stick with Nvidia A40—it balances performance and cost. Then, select 'Image Model' from the options. https://t.co/6oHc7tVpyC

@skirano - Pietro Schirano

Setup complete, let's move on to training 💪. Replicate offers various training options, but the terminal is my go-to. Head there and install the following: https://t.co/Kay98kMR9y

@skirano - Pietro Schirano

Next, set up your API keys. You'll find these in your account under 'API Token.' Then in terminal again run "export REPLICATE_API_TOKEN=<your token here>" https://t.co/vl0ly2CLkR

@skirano - Pietro Schirano

Back in the terminal, run the following command to start the training. You'll need your model's name, the link to your zipped images, and a unique keyword for your style (e.g., 'In the style of TOK') https://t.co/vNSyk5ejvL

@skirano - Pietro Schirano

The 'lora_lr' value controls training speed. Lower values make the model focus more on details. Experiment with 1e-4, 2e-4, or 4e-4.

@skirano - Pietro Schirano

Monitor the status at replicate.com/trainings. It usually takes ~10 mins and costs 90 cents And you are basically done! ✅

@skirano - Pietro Schirano

Hope you found this guide useful. For more insights on AI, from coding to image modeling and LLMs, hit that follow button!