I’ve created an AI double (fine-tuned a LLM) of my fiancée to prompt her. She hates it, but it’s great fun. You can achieve this in just a few minutes for yourself, your loved ones, your products, etc.
Need more photos for your LinkedIn posts? Pictures of your dog exploring Mars? Is your company spending too much on photography or professional editing for your products?
Follow this guide, have fun, and save money on your creative projects.
It’s super simple and can be done by anyone with 5 minutes (add an extra 25 minutes for the AI to complete its training).
Here’s how to do it, step by step:
1. Set Up Your Account
Go to Replicate and create an account. Add a payment method (you’ll spend less than €5 on the transaction).
Then go to this link.
2. Prepare Your Images
To make your model effective, the quality of your training images is essential.
- Number of images: Prepare at least 10 varied photos of yourself or your subject.
- Quality and context: Choose images that reflect how you intend to use them.
- Rename your files: Each file should include the subject’s name (e.g., Clement_01.jpg, Clement_02.jpg, etc.).
- Compress your images: Put all your photos into a folder, then compress the folder into a ZIP file.
3. Train Your Model
Using the provided Replicate link:
- Give your model a name (e.g., ClementGPT).
- Upload the ZIP file containing your images.
- Set a trigger keyword—this is the word or phrase you’ll use to generate images (e.g., Clement).
- Leave the other settings as default (they’re optimized for general use).
- Click on “Create a template”. The workout will take about 20 to 25 minutes.
4. Generate Images
Once you’ve finished training, you can use your model. Here’s how to generate images:
- Click Run Model.
- Enter a prompt (description of what you want to generate). Example: “Dynamic portrait of [déclencheur]” or
- Ajuster les paramètres :
- Aspect ratio: 16:9, 1:1…
- Number of images: Request several images at a time.
- Etc
- Click on “Launch”. In just a few seconds, your images will be ready. The results won’t be perfect. It depends on your prompts, uploaded images, and the model’s current capabilities.
5. Tips
- Elements from your training photos (backgrounds, accessories) may appear in the generated images. Choose simple, clean photos for the best results. Choose simple, clean photos for the best results.
- You can explore advanced parameters to improve consistency (such as setting a SEED for more stable results), or choose a longer training of your model (higher costs).
