I Was asked to show my process for creating loRA models from my Assorted girls collection where i make a lora from a single image so that's what I'm doing in this article.
1. getting the first photo.
The first step is selecting a photo to work with. Ideally, this image should feature a single character, shot from medium distance (at least from the crotch up). The details must be clear, so avoid blurry images. The character should face the viewer directly, and the pose should be close to neutral; avoid any closed-off poses that could complicate the training process.
Note: I have only used this method with fictional, realistic characters. I can't guarantee effectiveness with more cartoonish styles or with real people if you want an exact likeness.
2. getting more angles with A.I video
Now to get a good lora, as everyone knows, you'll need multiple photos—not just one. While you could copy and paste the same image, cropping each version differently (a method I used for the "Erin Model") results can be inconsistent.
For a more reliable model, I prefer using an AI video maker to generate additional angles of the subject. There are plenty of A.I. video makers now, personally i use KLING A.I. as it offers various options in its image-to-video converter and produces better results compared to the Hot mess i was using that is Luma.
Now when creating the video, you want to keep it simple. The goal is to capture multiple views of the subject without introducing major errors or oddities that could skew the model. A straightforward prompt works best, such as "girl looking at the viewer then turning away to a side profile view." You can generate multiple videos for even better results, provided they remain consistent and true to the original character.
the final video
3. Finishing up
After generating the video, import it into an editing software like DaVinci Resolve to extract the frames that you think look best.
From here, you can proceed with training the LoRA model as you normally would. For realistic pony models, I recommend using a model like Pony Realism, as I've found it tends to produce better and more accurate results than the standard base pony for this kind of model.
I’ve attached my LoRA settings for reference. I hope this guide has been helpful!