<h1 id="overview">Overview</h1>This LoRA is an SDXL-based character LoRA created to reproduce the <code>BACKGWA</code> character.This project was conducted for research purposes to protect the copyright, creative style, training rights, and intellectual property held by the original artist of this character. Instead of using images directly created by the artist as training data, this LoRA was trained only on an AI-generated synthetic image dataset.Through this approach, the project explores a way to build a character LoRA without directly using the original character images or the artist's work as training data. <h2 id="disclaimer">Disclaimer</h2><ul><li>This LoRA may be used freely for the user's own creative purposes. However, all copyright issues, legal responsibilities, and social or ethical consequences arising from any images, videos, or other outputs generated using this LoRA remain solely with the user.</li><li>Users must ensure that generated outputs do not infringe upon the rights, reputation, or creative works of the original artist, character rights holder, or any third party.</li><li>The creation and use of SFW and NSFW outputs are entirely at the user's own discretion and responsibility. The creator, distributor, and maintainer of this LoRA bear no responsibility or obligation regarding such outputs.</li><li>By using this LoRA, the user assumes full and exclusive responsibility for any legal disputes, social issues, infringement of third-party rights, defamation, or unlawful acts that may arise.</li><li>Users must comply with all applicable laws, platform rules, and ethical standards in their country or region when using this LoRA. Any consequences resulting from violations of such laws, rules, or standards shall be borne solely by the user.</li><li>Under no circumstances shall the creator, distributor, or maintainer of this LoRA be liable for any direct, indirect, incidental, special, or consequential damages arising from the use or inability to use this LoRA.</li><li>Redistribution of this LoRA by third parties is prohibited in principle. However, it may be permitted if the original creator is clearly credited and proper attribution is provided.</li></ul><h2 id="trigger-word">Trigger Word</h2>You can use the <code>BACKGWA</code> character by entering the trigger prompt below. This prompt includes the key visual features and default outfit elements required to generate the character.If you want to change specific details such as the outfit, expression, pose, background, or composition, you can keep the base trigger prompt and add or adjust the desired elements as needed.<pre><code>backgwa, solo, pale blue hair, cat ears, golden eyes, navy beret, white shirt, red ribbon, oversized sleeves, blue jacket</code></pre><h2 id="usage">Usage</h2>This LoRA was designed for use with SDXL-based model environments. For the most stable and consistent results, it is recommended to use it with the <a rel="ugc" href="https://civarchive.com/models/1760158/lumiere">LUMIERE-Q</a> model, though it may also be used with other SDXL-compatible checkpoints.<hr /><h1 id="training">Training</h1>This LoRA was trained without using any human-made artwork as training data.The dataset consists of 28 synthetic images generated with GPT Image 2. These images were created specifically for this LoRA training process and were not collected from, derived from, or reused from artwork produced by human artists.To prepare the initial image-generation prompts, the original character image was provided as an image input to the locally executed <a rel="ugc" href="https://huggingface.co/google/gemma-4-E4B-it">gemma-4-E4B-it</a> model. Predefined prompts and instructions were used to guide the model in producing text prompts capable of reconstructing the character from the provided reference image.Using those reconstructed prompts, synthetic character images were generated with GPT Image 2. When suitable high-quality outputs were obtained, some of them were also used as additional references for further image generation.The generated images were then labeled using GPT-5.5. These labels describe the visible character features, outfit elements, composition, and other relevant image details used during training. A total of 28 synthetic images were used for training. The LoRA was trained on the SDXL-based <a target="_blank" rel="ugc" href="https://civarchive.com/models/1760158/lumiere">LUMIERE-Q</a> model, with 10 repeats and 8 epochs at a resolution of <code>1024x1024</code>.

Files with SHA256: c106a4f76c407f3ec595abdf8308eea8ff19e70ca4f94cc60c3b352490f051b4