<h1 id="heading-17624">SciStyle</h1>v1 of SciStyle is a test model for a new image captioning pipeline I've been working on. The model was trained on a subset of 1k images of various styles/mediums. Surprised by the results for a model trained on only 1k images, I decided to release it here. The full model is currently being worked on.For more info on the image captioning pipeline, refer to my Discord thread linked bellow<hr /><h3 id="heading-107">Questions/Feedback/Updates?</h3><a target="_blank" rel="ugc" href="https://discord.com/channels/1010980909568245801/1073440375336882186">Visit my thread on the Unstable Diffusion Discord</a><hr /><h3 id="heading-17625">Info</h3>S&amp;DBase Model: Stable Diffusion v1.5Type: Experimental Fine-tuneClip: 1Medium: Multi-mediumCaption Style: Natural Language + Booru StyleDataset Size: Subset, 4k images out of 25k images + DnD datasetTraining Resolution: 768x768Difference from v1: More fantasy focused, additional training on a DnD dataset.<hr />V1Base Model: Stable Diffusion v1.5Type: Experimental Fine-tuneClip: 1Medium: Multi-mediumCaption Style: Natural Language + Booru StyleDataset Size: Subset, 1k images out of 25k imagesTraining Resolution: 768x768<hr />V2Base Model: Stable Diffusion v1.5Type: Experimental Fine-tuneClip: 1Medium: Multi-mediumCaption Style: Natural Language + Booru StyleDataset Size: Subset, 6.5k images out of 25k imagesTraining Resolution: 768x768Difference from v1: More species from various Sci-fi and fantasy universes.<hr /><h3 id="heading-17626">Features</h3><ol><li>Multi-medium: Capable of generating images from multiple art mediums, simply include the medium in the prompt.</li><li>Natural Language &amp; Booru: Accepts both natural language prompts and booru style prompts.</li><li>Extra Detail: Understands subtle details often skipped by SD models. Such as, number of objects/subjects in a scene, background information, color information for various parts of the image, atmosphere, ect.. (see my discord thread above for more info on how this is achieved.)</li><li>Flexible: Can easily be merged with other SD1.5 checkpoints / LoRAs</li></ol><hr /><h3 id="heading-17627">Usage</h3>Special Tokens:<ul><li><code>SciStyle</code>, can be used as a class token at the beginning of the prompt, but is not necessary.</li><li>Tag for various art mediums, i.e., <code>a comic book illustration of</code>, <code>90s anime screencap of</code> or, simply add the medium towards the end of the prompt; <code>comic book illustration</code>, <code>photorealistic</code>. These are just examples of tag placement. Feel free to experiment with other mediums</li></ul><hr /><h3 id="heading-17628">Recommended Settings</h3><h3 id="heading-17629">Sampler/Solver:</h3><ul><li>Euler a<ul><li>Steps: 20 - 32</li><li>CFG: 6 - 7.5</li></ul></li><li>DPM++ SDE Karras<ul><li>Steps: 30 - 40</li><li>CFG: 6 - 8.5</li></ul></li><li>DPM++ 2M SDE Karras<ul><li>Steps: 50+</li><li>CFG: 7 - 8</li></ul></li></ul>These are just recommendations.<h3 id="heading-17630">Hires Fix</h3><ul><li>4x-UltraSharp - <a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-UltraSharp">Link</a></li><li>Remacri - <a target="_blank" rel="ugc" href="https://openmodeldb.info/models/4x-Remacri">Link</a></li></ul>Settings for all ESRGAN models:<ul><li>Upscale by<ul><li>1.5 if resolution is &gt; 512x768</li><li>Don't exceed 2.0 (unless you have a beefy rig)</li></ul></li><li>Denoise Strength<ul><li>0.25 - 0.35</li></ul></li><li>Hires Steps<ul><li>If sampling steps &gt; 60,<ul><li>hires steps = half of sampling steps</li></ul></li><li>Otherwise, leave at 0</li></ul></li></ul><h3 id="heading-17631">Extensions</h3>ADetailer Download <a target="_blank" rel="ugc" href="https://github.com/Bing-su/adetailer">here</a>Neutral PromptDownload <a target="_blank" rel="ugc" href="https://github.com/ljleb/sd-webui-neutral-prompt">here</a>Read repo(s) Descriptions for usage guides<h3 id="heading-17632">Negative Embeddings</h3><ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/72437/baddream-unrealisticdream-negative-embeddings">BadDream</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/72437?modelVersionId=77173">UnrealisticDream</a></li></ul>Only if you want to remake one of the sample images. Personally, I would avoid using negative embeddings and instead use a simple negative prompt and then add+ or subtract- tokens per new idea. I only use them to speed-up inference during sample generation. That being said, other negative embeddings such as EasyNegative, ect.. are also fine to use with this model.<hr /><h3 id="heading-10511">Checkout my other models</h3>SDXL<ul><li>BAM - <a target="_blank" rel="ugc" href="https://civarchive.com/models/163139/boomer-art-model-bam">https://civarchive.com/models/163139/boomer-art-model-bam</a></li></ul><ul><li>LomoXL - <a target="_blank" rel="ugc" href="https://civarchive.com/models/156743/lomoxl">https://civarchive.com/models/156743/lomoxl</a></li></ul>SD1.5<ul><li>Revival - <a target="_blank" rel="ugc" href="https://civarchive.com/models/149738/revival">https://civarchive.com/models/149738/revival</a></li></ul><ul><li>Doomer Boomer - <a target="_blank" rel="ugc" href="https://civarchive.com/models/118247/doomer-boomer">https://civarchive.com/models/118247/doomer-boomer</a></li><li>Lomostyle - <a target="_blank" rel="ugc" href="https://civarchive.com/models/109923/lomostyle">https://civarchive.com/models/109923/lomostyle</a></li></ul><ul><li>Another Damn Art Model (ADAM) - <a target="_blank" rel="ugc" href="https://civarchive.com/models/104898/another-damn-art-model-adam">https://civarchive.com/models/104898/another-damn-art-model-adam</a></li></ul><ul><li>Based Model - <a target="_blank" rel="ugc" href="https://civarchive.com/models/83991?modelVersionId=89262">https://civarchive.com/models/83991?modelVersionId=89262</a></li><li>Electric Eden - <a target="_blank" rel="ugc" href="https://civarchive.com/models/64355/electric-eden">https://civarchive.com/models/64355/electric-eden</a></li></ul><ul><li>Cine Diffusion - <a target="_blank" rel="ugc" href="https://civarchive.com/models/50000/cine-diffusion">https://civarchive.com/models/50000/cine-diffusion</a></li><li>ProjectAIO - <a target="_blank" rel="ugc" href="https://civarchive.com/models/18428/project-aio">https://civarchive.com/models/18428/project-aio</a></li></ul><ul><li>Eris - <a target="_blank" rel="ugc" href="https://civarchive.com/models/21952/eris">https://civarchive.com/models/21952/eris</a></li></ul><ul><li>WonderMix - <a target="_blank" rel="ugc" href="https://civarchive.com/models/15666/wondermix">https://civarchive.com/models/15666/wondermix</a></li></ul><ul><li>Refined - <a target="_blank" rel="ugc" href="https://civarchive.com/models/8392/refined">https://civarchive.com/models/8392/refined</a></li></ul><ul><li>Experience - <a target="_blank" rel="ugc" href="https://civarchive.com/models/5952/experience">https://civarchive.com/models/5952/experience</a></li><li>Elegance - <a target="_blank" rel="ugc" href="https://civarchive.com/models/5564/elegance">https://civarchive.com/models/5564/elegance</a></li></ul><ul><li>Clarity - <a target="_blank" rel="ugc" href="https://civarchive.com/models/5062/clarity">https://civarchive.com/models/5062/clarity</a></li><li>VisionGen - Realism Reborn -<a target="_blank" rel="ugc" href="https://civarchive.com/models/4834/visiongen-realism">https://civarchive.com/models/4834/visiongen-realism</a></li></ul>LoRA<ul><li>Pant Pull Down - <a target="_blank" rel="ugc" href="https://civarchive.com/models/11126/pant-pull-down-lora">https://civarchive.com/models/11126/pant-pull-down-lora</a></li></ul>

SciStyle

SciStyle-v2<ul><li>More races/species from various Sci-fi and Fantasy universes.</li><li>Started fine-tuning photo manipulation into the model for more abstract generations. This idea has since been abandoned and will be moved to its own dedicated photo manipulation model.</li></ul>

Files with SHA256: 002ecbc866d99b0a916eca2eaa86989cef81f3cdc8605d33c9d3b17b614e84fd