<h1 id="read-description-0p2cjtfv4">Read Description</h1>Note: Clarity XL is currently in BETA.Fine-Tuning is still in progress.Like Photorealism? Checkout my latest SDXL Fine-Tune: <a target="_blank" rel="ugc" href="https://civarchive.com/models/617652/natvis-natural-vision">NatViS</a>.<hr /><h3 id="changelog-4bx3muv7k">Changelog</h3>8/26/24 ClarityXL v2.0 Lightning 8step<ul><li>Released 8step lightning version of ClarityXL v2.0, by request. Make sure you read the About this version for more info.<ul><li>Note: Use a lower CFG (1.5 - 2.5) if colors look washed out. I made the mistake of setting it to high in the sample images.</li></ul></li></ul>————8/12/24 ClarityXL v2.0<ul><li>Released v2.0 of ClarityXL. Read About this version to see what's new.</li></ul><hr /><h3 id="buy-me-a-coffee-bz9lgfevl">Buy me a coffee ❤</h3><a target="_blank" rel="ugc" href="https://ko-fi.com/ndimensional">https://ko-fi.com/ndimensional</a>I’ve never been a fan of e-begging, however SDXL fine-tunes at this scale are becoming expensive to tune. So I will begrudgingly ask; if you like what I do and would like to support my models. Consider donating on <a target="_blank" rel="ugc" href="https://ko-fi.com/ndimensional">Ko-Fi </a>💗 I will be begin posting updates, answering questions, taking feedback, and releasing early access (NOT EXCLUSIVE) models to supporters.All donations will be used to fund the creation of new Stable Diffusion fine-tunes and open-source AI tools.<hr /><h2 id="about-vmfh0bg2f">About</h2>Continuing from the <a target="_blank" rel="ugc" href="https://civarchive.com/models/5062/clarity">original Clarity model for SD1.5</a>, Clarity XL is an attempt to recreate and expand the original models capabilities within the more complex architecture of SDXL.<h3 id="differences-between-clarity-sd1.5-and-clarity-xl-gfkcgi3wp">Differences between Clarity SD1.5 and Clarity XL</h3>Currently, Clarity XL focuses purely on photorealism. This is intentional to build the foundation which will be expanded upon in future releases. That's not to say Clarity XL will ever be a general purpose model. It will always have a bias towards photorealism. Future releases will add more complex photorealistic/cinematic scene capabilities.Improvements<ul><li>Emphasis on authentic (non touched-up) photorealism.</li><li>Higher Image Fidelity.</li><li>Prompt Adherence: How well the model follows your prompt.<ul><li>Excluding concepts that the model was not trained on</li></ul></li><li>Improved skin textures</li><li>Overall improvements to aesthetics.</li><li>Video Game / Movie character recognition.<ul><li>Including worlds spaces, landscapes, settings, ect..</li></ul></li><li>Prompt how you want: Accepts natural language prompts, comma delimited-lists, a hybrid of the two. In addition, prompts can be as short or long as you'd like.</li></ul>Limitations<ul><li>Complex scenes, such as <code>firing lighting bolts from hand, erupting in a cacophony of bright blue sparkling arcs.</code></li><li>Multi-medium generation: The model is currently grounded in photorealism and cinematography.</li></ul><h3 id="model-details-vzmi20tsm">Model Details</h3><ul><li>Base Model: <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0">Stable Diffusion XL v1.0</a><ul><li>Since Clarity XL v1 is a mid-training epoch. I merged the Epoch with an unreleased fine-tune update for <a target="_blank" rel="ugc" href="https://civarchive.com/models/156743/lomoxl">LomoXL</a>. I used a modified version of the <a target="_blank" rel="ugc" href="https://arxiv.org/pdf/2311.03099">DARE </a>merging method to preserve the original weight matrix of the base epoch. This will not be needed in later releases.</li></ul></li><li>Data: Quality was a priority when creating the dataset. All image-caption pairs were cleansed through multiple iterations to ensure only high quality data was used for tuning.<ul><li>Captions: Captions were written by my MLLM captioning system, verified via <a target="_blank" rel="ugc" href="https://github.com/IDEA-Research/GroundingDINO">GroundingDINO </a>+ a Reasoning Engine + NLP<ul><li>Captions were written in a natural language format. Though, SDXL's text-encoders make it possible to write prompts in multiple prompting styles.</li></ul></li></ul></li><li>VAE: <a target="_blank" rel="ugc" href="https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl_vae.safetensors">sdxl-vae-fp16-fix</a></li><li>Aspect Ratio: From training data, any of the typical aspect ratios for SDXL will work.<ul><li>1344x768 (16:9) — Cinematic Film Stills</li><li>1536x640 (21:9) — Ultrawide Cinematic Film Stills</li><li>1152x896 (4:3) — Fullscreen</li><li>1216x832 (3:2) — Mobile landscape</li><li>1024x1024 (1:1) — Square</li><li>1024x704 (11:16)</li><li>768x1344 (9:16) — Tall (Instagram stories / snapchat)</li><li>896x1152 (3:4)</li><li>832x1216 (2:3) — Mobile Portrait</li><li>704x1024 (16:11)</li></ul></li></ul>🤗<a target="_blank" rel="ugc" href="https://huggingface.co/nDimensional/Clarity-SDXL">Huggingface Repo</a><hr /><h2 id="changelog-6zb2o8g3n">Changelog</h2>5/23/24 Clarity XL v1.0:<ul><li>Initial Release</li></ul><hr /><h2 id="checkout-my-other-models-xtxfi9x8z">Checkout my other models</h2>SDXL Checkpoints: <a target="_blank" rel="ugc" href="https://civarchive.com/collections/966964">https://civarchive.com/collections/966964</a>SDXL LoRAs: <a target="_blank" rel="ugc" href="https://civarchive.com/collections/966969">https://civarchive.com/collections/966969</a>40K Series: <a target="_blank" rel="ugc" href="https://civarchive.com/collections/956187">https://civarchive.com/collections/956187</a>SD1.5 Checkpoints: <a target="_blank" rel="ugc" href="https://civarchive.com/collections/966974">https://civarchive.com/collections/966974</a>SD1.5 LoRAs: <a target="_blank" rel="ugc" href="https://civarchive.com/collections/966972">https://civarchive.com/collections/966972</a><hr />Run Clarity XL on Tensor Art: <a target="_blank" rel="ugc" href="https://tensor.art/models/740350571376126636?source_id=nj2-r1nnnUO3ovUiaHf19Bgn">https://tensor.art/models/740350571376126636?source_id=nj2-r1nnnUO3ovUiaHf19Bgn</a>

Clarity XL

What's new in V2?<ul><li>Improvements to prompt adherence. i.e., how closely the generation follows your prompt. <ul><li>Note: This might mean some prompts from v1.0 might need to be slightly reworded. As v1.0 had a tendency to drop certain tokens. Sometimes, this resulted in unexpected but sometimes pleasant results. These results can still be achieved with a little bit of tweaking of the prompt.</li></ul></li><li>Improvements to NSFW generation.</li><li>Likewise, the model should have less of a tendency to generate NSFW when not prompted to do such.</li><li>Additional training on Natural Language captions, using subset of <a rel="ugc" href="https://civitai.com/models/617652/natvis-natural-vision">NatViS </a>dataset.</li></ul>What's Next?<ul><li>Continued Improvements to prompt adherence.</li><li>Inclusion of more Natural Language captions.</li></ul>

Files with SHA256: 06f07f466db3f50085a9f893bcaa9d30c33889033eb8641cdbfd52ab245b02a3