This LoRA is my first attempt at a LoRA of Windows Folder Design.Civitai's default settings made this model very small and made it not keep the consistency of its dataset. In about 8 hours, there should be a better, more consistent model up. The dataset consists of 215 images of consistent Windows folder designs captioned by gpt4-o using <a target="_blank" rel="ugc" href="https://sd-caption-helper.vercel.app/">sd-caption-helper</a> (<a target="_blank" rel="ugc" href="https://sd-caption-helper.vercel.app/">https://sd-caption-helper.vercel.app/</a>). The captions are full sentences to match flux. But I would imagine that "folder icon" would help as a trigger word. Taking suggestions for LoRAs for SD 1.5, SDXL and Flux. If you have something, a style, a character or a concept that could be interesting to make a LoRA of please give me ideas. ComfyUI workflow: <a target="_blank" rel="ugc" href="https://openart.ai/workflows/ailab/flux-lora-simple/gErKNyjsnetZpEgaMTrV">https://openart.ai/workflows/ailab/flux-lora-simple/gErKNyjsnetZpEgaMTrV</a> Changelog:Version 1.5: Improved consistency through higher rank size Workflow 1. From over 2.000 images, I've picked images I thought would enhance the LoRA the most. 2. I've used <a target="_blank" rel="ugc" href="https://sd-caption-helper.vercel.app/">SD-caption-helper</a> to caption the images. This is the prompt I've used: Write detailed full sentence caption for this image while keeping it as short as possible avoiding all unnecessary words like "the". Mention the Colors, design, logo company name and mention all text depicted in the image in quotes. The caption should be between 40 and 75 words long. Don't mention the "image" just get straight into the description like "A ...". Avoid overly flowery language. Give me the description only. No unnecessary mention of you task. Keep it plain and simple and short. Don't wrap the whole prompt in any quotes or similar. Only describe what can be seen and not what can be assumed unless its sure. Make full sentences as simple as possible. 3. I've used Civitai Flux training for version 1.0. <a target="_blank" rel="ugc" href="https://github.com/ostris/ai-toolkit">ai-toolkit </a>for version 1.5PS: If you want the dataset please contact me. I just don't want to get CivitAI in copyright trouble.

FolderDesign

215 images, Civitai Training, gpt4-o caption

Files with SHA256: 841f760cb722db2fdfe485f410d083f0767d1ec68c2a7e5f2ee331df360e43d9