Create README.md

README.md

@@ -0,0 +1,45 @@
+---
+license: other
+tags:
+- flux
+- lora
+- text-to-image
+- image-to-image
+base_model: black-forest-labs/FLUX.1-dev
+instance_prompt: lisa
+widget:
+- text: "lisa wearing a black dress and black platform teva sandals at the beach, high resolution"
+  output:
+    url: "sample_output.png"
+---
+
+# Lisa FLUX LoRA (sscorzo/ls)
+
+This is a LoRA trained on the FLUX.1-dev model using Replicate. It is optimized to capture the likeness of **Lisa** across various settings and outfits, with a specific focus on style consistency.
+
+## Training Details
+The model was trained with the following core configurations:
+* **Trigger Word:** `lisa`
+* **Secondary Concepts:** The model was also exposed to descriptive tags including `lisa wearing tevas`, `lisa wearing a black dress`, and `lisa at the beach`.
+* **Network Rank/Alpha:** 16 / 16
+* **Learning Rate:** 0.0004
+* **Steps:** 1000
+* **Optimizer:** AdamW 8-bit
+* **Training Resolution:** Multi-scale (512, 768, 1024)
+
+## Usage
+
+### Recommended Prompting
+To activate the likeness, include the trigger word **lisa** in your prompt. For best results with the footwear styles included in the training data, use specific descriptive phrases:
+
+> **Prompt:** `lisa wearing a black dress and black platform teva sandals, standing on a pier at sunset, 8k resolution.`
+
+### Recommended Settings
+* **Guidance Scale:** 3.5
+* **Inference Steps:** 28
+* **LoRA Weight:** 0.7 - 1.0 (Adjust based on how strong you want the likeness to be)
+
+## Files in this Repository
+* `lisa_flux_v1.safetensors`: The primary LoRA weight file.
+* `config.yaml`: The full training configuration file used on Replicate.
+* `captions/`: A folder containing the image-specific text descriptions used during the training process.