DeepMode Blogs | Blogs about AI character generation

What is Lora?

Simply put, Lora is a tag that looks like this: <lora:lora_name:1> that you add to your prompt which was trained on a specific set of photos — that helps the AI understand better the desired action you want. LoRA, or Low-Rank Adaptation, is a powerful technique in AI image generation that allows users to fine-tune models for specific styles, characters, or concepts without extensive retraining. In the context of platforms like deepmode.ai, LoRAs serve as specialized add-ons that can significantly enhance the capabilities of base models.

Here's an expanded guide on using NSFW LoRAs in deepmode.ai

Understanding LoRAs

LoRAs are small AI models trained on specific datasets to capture particular attributes, styles, or concepts. They can range from artistic styles to character designs, and even to more specific elements like clothing or accessories. When applied to a base model, LoRAs influence the generation process, allowing for more targeted and customized outputs.

Finding DeepMode's LoRAs

We keep adding new Loras all the time based on suggestions from our community on discord. If you want to add a specific Lora just ask for it on our discord channel. You can view our existing NSFW Loras HERE or browse our community Lora collection by clicking on the "+ Lora / Style" button below the prompt.

Using LoRAs in Image Generation

When you browse the community Lora Gallery you will see a lot of Lora styles. Clicking on a Lora will open a preference window with the Lora Tag along side with Trigger words to activate the Lora. You can select keywords by clicking on the ones you want to add, and those will be added to your prompt. Notice that not all Loras have trigger words and that's fine. Some Loras don't need Trigger words, and simply including the Lora tag will do the job.

Optional: Choose the trigger words that activate the Lora. Press "Add to Prompt" to add the tag to your prompt.

Using LoRAs Weights in Image Generation

After the Lora tag was added to your prompt you will notice something like this: <lora:lora_name:0.8>. This is the Tag of the Lora name followed by a number (set to 0.8 by default). You can change the value strength of your lora by changing the number from 0.8 to anywhere between 0.1 and 1. You can also go over 1 but be warned that it's not recommended as you will loose the similiarity to you subject. An average value of 0.8 is usually a good place to start. Keep in mind, the use of Loras in your prompt might change the similarity of your model so you need to adjust it the right way to get a good result.

Advanced Techniques

1. Negative LoRAs: Some platforms support negative LoRA weights, allowing you to specify styles or elements you want to avoid in the generation.

2. LoRA Stacking: Layer multiple LoRAs with different strengths to create complex results.

Tips for Effective LoRA Use

• Start Simple: Begin with a single LoRA and gradually experiment with combinations as you become more comfortable.

• Read LoRA Descriptions: Pay attention to the creator's notes on trigger words and recommended usage.

• Iterate and Refine: Don't be afraid to generate multiple images, tweaking your prompt and LoRA settings each time.

• Community Resources: Engage with the deepmode.ai community on discord to discover new LoRAs and learn from others' experiences.

By mastering the use of LoRAs in deepmode.ai, you can significantly expand your creative possibilities, producing highly customized and unique AI-generated images that align closely with your artistic vision.