<h1 id="v-pred-branch-of-cyberfalafel">V-pred branch of Cyberfalafel</h1>Completely new model type for me, but it's work and work good.Model have medium cfg sensitivity. CFG 4 or 5 is recommended. Recommended vae xlVAEC_f22 or f231. MaHiRo is also recommended if cfg scale on 4-4.5.<h2 id="recommend-sampler:">Recommend sampler:</h2>DPM++ 2m. And you can use DPM ++2M SDE by the way. Euler A. Euler. DDIM.<h2 id="schedule-type:">Schedule type:</h2>Beta. SGM uniform/normal/simple/uniform. Polyexponential. Exponential.For upscale process i recommend DPM++2 m SDE + SGM uniform. For inpainting DPM++ 2m + SGM uniform with 0.25-0.3 denoise and 20-30 steps (if you use forge neo/reforge/forge classic - use formula: steps*denoise = 20-30. 90 steps work fine). Model can handle large tensor resolution (1536x1536), and i can recommend this on general use if you making composition like portrait, full body standing or something wide shot. Thanks to the new predictor, the model handle with this many times better than the EPS one. Generic positive: very as2, masterpiece, absurdres, best quality, very awa, very aesthetic, year 2024, newest, ultra-detailed, amazing quality. Generic negative: (patreon username, patreon logo, artist logo, twitter username, signature, username, logo, watermark, copyright name, copyright symbol, web address, twitter logo, gumroad username, text, artist name, gumroad logo, artist logo, logo, blurry, blurry background, blurry foreground, blurry edges, blurry vision, bokeh, depth of field:1.2), green theme, red theme, yellow theme, blue theme, worst quality, worst aesthetic, bad quality, low quality, (bad hands:1.2), censored, bar censor, mosaic censoring, artistic error, bad proportions, bad perspective, wrong hand, bad reflections, too many fingers, too many hands, too many legs, wrong feet, bad feet, extra digits, missing fingers,bad anatomy,anime screenshot, anime coloring, abstract, ai-generated, lossy-lossless, jpeg artifacts, scan artifacts,===========================================================I made cyberfix for falafel mix by <a target="_blank" rel="ugc" href="https://civarchive.com/user/DS27SD">ds27sd</a>This checkpoint has the same reason and goals as the original cyberfix, as well as the creation method. Many thanks to the original author.Recommended settings: CFG 5, sampler euler A AYS, 28-35 steps.<h1 id="original-description">Original description</h1>Original model: <a target="_blank" rel="ugc" href="https://civarchive.com/models/913998/noobaicyberfix?modelVersionId=1022962">https://civarchive.com/models/913998/noobaicyberfix?modelVersionId=1022962</a><h1 id="noobaicyberfix">NoobaiCyberFix</h1>Purpose of this mergeThe purpose of this model is to fix the anatomy problems of model <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294/noobai-xl-nai-xl">NoobAI</a>, while keeping 100% of its knowledge and style. NoobAI-Epsilon075 is a great model, it knows almost all characters and styles from Danbooru, but it has some anatomy issues and this merge completely solves this problem. If you use NoobAI-Epsilon075 for SFW purposes, you will hardly notice any difference between this merge and the original model, because the goal was to keep all the knowledge of NoobAI-Epsilon075, and visually this model is almost indistinguishable from NoobAI-Epsilon075. But if you are a man of culture and like Pony models, you will definitely like this one, because this merge gives the model a very good knowledge of anatomy, almost at Pony levelRecommended negative prompt<pre><code>lowres, (worst quality, low quality, bad anatomy, bad hands:1.3), abstract, signature</code></pre>Recipe of the merge1 Create temporary modelTempModel = <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294/noobai-xl-nai-xl">NoobAI-Epsilon075</a> + <a target="_blank" rel="ugc" href="https://civarchive.com/models/312530/cyberrealistic-xl">cyberrealisticXL_v4</a> - <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0">stable-diffusion-xl-base-1.0</a>2 Remove CLIP from TempModel and replace it by CLIP from <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294/noobai-xl-nai-xl">NoobAI-Epsilon075</a>I did this step in ComfyUI. Here is the workflow <a target="_blank" rel="ugc" href="https://civarchive.com/models/991574?modelVersionId=1110897">https://civarchive.com/models/991574?modelVersionId=1110897</a>Logic behind this mergecyberrealisticXL_v31 is a general purpose model without any specific style, just like stable-diffusion-xl-base-1.0, and so if you replace the stable-diffusion-xl-base-1.0 weights with cyberrealisticXL_v31 weights in any model, it shouldn't change the model's style and its knowledge much. But at the same time, cyberrealisticXL_v31 has much better anatomy knowledge than sd-xl-1.0, and because of this, it should greatly improve the anatomy knowledge of any model, and in the case of NoobaiCyberFix, it worked exactly as I expected. In theory, you can use this recipe to improve the anatomy knowledge of any model that needs itBig thanks to the creators of these models!<a target="_blank" rel="ugc" href="https://civarchive.com/models/833294/noobai-xl-nai-xl">NoobAI</a><a target="_blank" rel="ugc" href="https://civarchive.com/models/795765/illustrious-xl">Illustrious-XL</a><a target="_blank" rel="ugc" href="https://civarchive.com/models/312530/cyberrealistic-xl">cyberrealisticXL_v4</a><a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0">stable-diffusion-xl-base-1.0</a>

Cyberfalafel - falafel cyber fix

An almost complete reinterpretation of the original recipe version 2.2. New models, new mixing principles, and it's now officially v-pred. This is my first experience working with this type of predictor, but personally I was extremely satisfied with the outcome of the work. The model retained all the advantages of the original EPS version, while acquiring all the advantages of the v-pred. All preview image made without any upscale and with minimum inpainting!Model have medium cfg sensitivity. CFG 4 or 5 is recommended. Recommended vae xlVAEC_f22 or f231. MaHiRo is also recommended.Recommend sampler - DPM++ 2m (or euler) for generating process. Scheduler - polyexponential. dpm++ 2m for inpaint (sgm uniform or polyexponential) and upscale. 28-35 steps.Model can handle large tensor resolution (1536x1536), and i can recommend this on general use if you making composition like portrait, full body standing or something wide shot. Thanks to the new predictor, the model copes with this many times better than the previous one.

Files with SHA256: 02c3054ba1e15a05b5540ba8cdedfd20e6bc3582c2af5561263b52e4df898b4c