--- tags: - text-to-image - stable-diffusion - lora - diffusers - template:sd-lora widget: - text: >- YFG-SpyWorld50, portrait, looking up, solo, upper body, detailed background, detailed face, CyberpunkAI theme, inventor, safety-goggles, shirt, innovation, advanced technology, fantastical science lab in background, chemistry, medicine, [lab mask:gas mask:0.2], test chamber, , spy-theme, 50s, (Masterpiece:1.3) (best quality:1.2) (high quality:1.1) parameters: negative_prompt: >- easynegative, watermark, low quality, ugly, blurry, lowres, bad hand, blurry eye output: url: images/ComfyUI__00104_.png base_model: stabilityai/stable-diffusion-3-medium-diffusers instance_prompt: YFG-SpyWorld50 --- # YFG-SpyWorld50 ## Model description This is an experimental LoRA trained using a non-production branch of OneTrainer. Uploading to share with anyone interested in this topic of training SD3 and for me to seek help from those who are more expert in how to better or perfect the results. Used a base script created by [@iceycold](https://huggingface.co/iceycold) (check out his great sd3Ballz LoRA) The data set came from CivitAI which was shared by [mnemic](https://civitai.com/user/mnemic); [Spy World 50s](https://civitai.com/models/150842/spy-world-1950s-surveillance-paranoia-style) which he used to create his SD1.5 LoRA. I am just reusing his data set to create a SD3 version. Here are some examples so far. Positive Prompt ``` YFG-SpyWorld50, coffee machine, spy-theme, 50s, (Masterpiece:1.3) (best quality:1.2) (high quality:1.1) ``` Negative Prompt ``` easynegative, watermark, low quality, ugly, blurry, lowres, bad hand, blurry eye ``` seed: 158464701716210 steps: 30 cfg: 3.5 Some samples LoRA bypassed ![image.png](https://cdn-uploads.huggingface.co/production/uploads/65dab6b80af7e21ba46347a4/3EuqRqMCT9OW2_coxafO-.png) LoRA enabled at 40% ![image.png](https://cdn-uploads.huggingface.co/production/uploads/65dab6b80af7e21ba46347a4/70BqOuYdLM9OYM9qBWLKX.png) LoRA at 100% ![image.png](https://cdn-uploads.huggingface.co/production/uploads/65dab6b80af7e21ba46347a4/KCLIQ7HIugzFOCp7Rxdop.png) The rest of the samples are at 100% for CLIP and LoRA strength ![image.png](https://cdn-uploads.huggingface.co/production/uploads/65dab6b80af7e21ba46347a4/eDYkPBibHb6V8IRFNeLbb.png) ![image.png](https://cdn-uploads.huggingface.co/production/uploads/65dab6b80af7e21ba46347a4/CdBYr506qbNbJyerNgMlj.png) ![image.png](https://cdn-uploads.huggingface.co/production/uploads/65dab6b80af7e21ba46347a4/ZS_dh0DcLsNDSI6l9AcwJ.png) ![image.png](https://cdn-uploads.huggingface.co/production/uploads/65dab6b80af7e21ba46347a4/9_NRGSFD2zZaU49V1ouU9.png) ![image.png](https://cdn-uploads.huggingface.co/production/uploads/65dab6b80af7e21ba46347a4/6isjUihHGgfMTIPEAa_D1.png) ![image.png](https://cdn-uploads.huggingface.co/production/uploads/65dab6b80af7e21ba46347a4/mf83WIugxRTX_D-0ox4nt.png) ## Trigger words You should use `YFG-SpyWorld50` to trigger the image generation. ## Download model Weights for this model are available in Safetensors format. [Download](/gonzalu/YFG-SpyWorld50/tree/main) them in the Files & versions tab.