If you’re re-training a style model:
-
You’ll be prompted to “describe the style”
-
This description will be used as part of the prompt for generations, and is essential for guiding your model’s style.
-
Keep it short and focused.
-
Examples: “impressionist painting” “cyberpunk digital art” “minimalist line drawing”
-
You will be able to tweak it later in the model settings or on the create page.
If you’re re-training a character/object/face model:
- no need for any extra steps, just click “re-train” and we’ll handle the rest
Creating with a re-trained model:
- When you create with a re-trained model, you might find that your old prompt formula, which used to work perfectly, now gives you some unexpected results.
- You may need to adjust the prompt formula to achieve the desired outcomes. This might take some testing and trial and error.
- Try adding or removing certain words based on the results.
- For example, if the colors are too bold, remove descriptors like “vivid” or “colorful.” If the style looks photorealistic instead of painterly, try adding words like “painting” or “brushstrokes.”
Read an example of how I adjusted the prompt formula for one of the models →
If you still don’t like the new model…
- Most likely, the training images that used to work with Stability (the old technology) don’t do so well with FLUX. You may want to re-curate a training set.
- For example, if you notice the generated images consistently emphasize a certain trait that shouldn’t be so important (like, a hand gesture, or a piece of clothing), you may want to remove the image with that trait and train another model.
