0. About Different Versions:-Original SDXL - Works as intended, correct CLIP modules with different prompt boxes. Unlike the previous SD 1. 0 Refiner. You can type in text tokens but it won’t work as well. Run update-v3. ago. Step 3: Download the SDXL control models. SDXL two staged denoising workflow. I've been having a blast experimenting with SDXL lately. In this guide, we'll show you how to use the SDXL v1. The the base model seem to be tuned to start from nothing, then to get an image. 5 Model works as Refiner. 9 and Stable Diffusion 1. 0 base and refiner and two others to upscale to 2048px. When trying to execute, it refers to the missing file "sd_xl_refiner_0. The refiner is trained specifically to do the last 20% of the timesteps so the idea was to not waste time by. To test the upcoming AP Workflow 6. ago. 0. ControlNet Workflow. Examples. I describe my idea in one of the post and Apprehensive_Sky892 showed me it's arleady working in ComfyUI. webui gradio sd stable-diffusion stablediffusion stable-diffusion-webui sdxl Updated Oct 28 , 2023. update ComyUI. 1: Support for Fine-Tuned SDXL models that don’t require the Refiner. 0 設定. I trained a LoRA model of myself using the SDXL 1. 33. 0s, apply half (): 2. 8s)Chief of Research. Use at your own risk. T2I-Adapter aligns internal knowledge in T2I models with external control signals. I'ma try to get a background fix workflow goin, this blurry shit is starting to bother me. safetensors and sd_xl_base_0. Is the best balanced I could find between image size (1024x720), models, steps (10+5 refiner), samplers/schedulers, so we can use SDXL on our laptops without those expensive/bulky desktop GPUs. So I think that the settings may be different for what you are trying to achieve. Installing ControlNet for Stable Diffusion XL on Windows or Mac. Do you have ComfyUI manager. refinerはかなりのVRAMを消費するようです。. RunPod ComfyUI Auto Installer With SDXL Auto Install Including Refiner. I tried Fooocus yesterday and I was getting 42+ seconds for a 'quick' generation (30 steps). Then refresh the browser (I lie, I just rename every new latent to the same filename e. The refiner model works, as the name suggests, a method of refining your images for better quality. Set the base ratio to 1. 5 model (directory: models/checkpoints) Install your loras (directory: models/loras) Restart. Ive had some success using SDXL base as my initial image generator and then going entirely 1. Thanks for your work, i'm well into A1111 but new to ComfyUI, is there any chance you will create an img2img workflow?Drawing inspiration from StableDiffusionWebUI, ComfyUI, and Midjourney’s prompt-only approach to image generation, Fooocus is a redesigned version of Stable Diffusion that centers around prompt usage, automatically handling other settings. 2占最多，比SDXL 1. We name the file “canny-sdxl-1. While the normal text encoders are not "bad", you can get better results if using the special encoders. 25:01 How to install and use ComfyUI on a free. If this is. The solution to that is ComfyUI, which could be viewed as a programming method as much as it is a front end. 9. 0 and upscalers. A (simple) function to print in the terminal the. 9版本的base model,refiner model. With resolution 1080x720 and specific samplers/schedulers, I managed to get a good balanced and a good image quality, first image with base model not very high. 17:18 How to enable back nodes. How to install ComfyUI. With Automatic1111 and SD Next i only got errors, even with -lowvram. Put into ComfyUImodelsvaeSDXL and ComfyUImodelsvaeSD15). You will need ComfyUI and some custom nodes from here and here . Prerequisites. Despite relatively low 0. Is the best balanced I could find between image size (1024x720), models, steps (10+5 refiner), samplers/schedulers, so we can use SDXL on our laptops without those expensive/bulky desktop GPUs. refinerモデルを正式にサポートしている. -Drag and Drop *. Also, you could use the standard image resize node (with lanczos or whatever it is called) and pipe that latent into sdxl then refiner. It will crash eventually - possibly RAM but doesn't take the VM with it - but as a comparison that one "works". The Stability AI team takes great pride in introducing SDXL 1. Drag & drop the . When you define the total number of diffusion steps you want the system to perform, the workflow will automatically allocate a certain number of those steps to each model, according to the refiner_start. It isn't strictly necessary, but it can improve the results you get from SDXL, and it is easy to flip on and off. それ以外. 0 through an intuitive visual workflow builder. 5 tiled render. WAS Node Suite. Using the SDXL Refiner in AUTOMATIC1111. Special thanks to @WinstonWoof and @Danamir for their contributions! ; SDXL Prompt Styler: Minor changes to output names and printed log prompt. for - SDXL. 5. Fine-tuned SDXL (or just the SDXL Base) All images are generated just with the SDXL Base model or a fine-tuned SDXL model that requires no Refiner. Model Description: This is a model that can be used to generate and modify images based on text prompts. Yup, all images generated in the main ComfyUI frontend have the workflow embedded into the image like that (right now anything that uses the ComfyUI API doesn't have that, though). The only important thing is that for optimal performance the resolution should be set to 1024x1024 or other resolutions with the same amount of pixels but a different aspect ratio. update ComyUI. The creator of ComfyUI and I are working on releasing an officially endorsed SDXL workflow that uses far less steps, and gives amazing results such as the ones I am posting below Also, I would like to note you are not using the normal text encoders and not the specialty text encoders for base or for the refiner, which can also hinder results. Fooocus uses its own advanced k-diffusion sampling that ensures seamless, native, and continuous swap in a refiner setup. The two-model setup that SDXL uses has the base model is good at generating original images from 100% noise, and the refiner is good at adding detail at 0. I did extensive testing and found that at 13/7, the base does the heavy lifting on the low-frequency information, and the refiner handles the high-frequency information, and neither of them interferes with the other's specialtysdxl_v1. 5 fine-tuned model: SDXL Base + SD 1. He puts out marvelous Comfyui stuff but with a paid Patreon and Youtube plan. In ComfyUI this can be accomplished with the output of one KSampler node (using SDXL base) leading directly into the input of another KSampler node (using. Stability is proud to announce the release of SDXL 1. Increasing the sampling steps might increase the output quality; however. 2. safetensors and sd_xl_base_0. SDXL-ComfyUI-workflows This repository contains a handful of SDXL workflows I use, make sure to check the usefull links as some of these models, and/or plugins are. . x for ComfyUI ; Table of Content ; Version 4. In my ComfyUI workflow, I first use the base model to generate the image and then pass it. 12:53 How to use SDXL LoRA models with Automatic1111 Web UI. 0 Refiner model. thibaud_xl_openpose also. Sample workflow for ComfyUI below - picking up pixels from SD 1. Basically, it starts generating the image with the Base model and finishes it off with the Refiner model. Locate this file, then follow the following path: Is there an explanation for how to use the refiner in ComfyUI? You can just use someone elses workflow of 0. 5 base model vs later iterations. 9 and Stable Diffusion 1. Inpainting a cat with the v2 inpainting model: . safetensor and the Refiner if you want it should be enough. It didn't work out. Workflow for ComfyUI and SDXL 1. How To Use Stable Diffusion XL 1. Stable Diffusion XL comes with a Base model / checkpoint plus a Refiner. SDXLの導入〜Refiner拡張導入のやり方をシェアします。 ①SDフォルダを丸ごとコピーし、コピー先を「SDXL」などに変更 今回の解説はすでにローカルでStable Diffusionを起動したことがある人向けです。 ローカルにStable Diffusionをインストールしたことが無い方は以下のURLが環境構築の参考になります。AP Workflow 3. This node is explicitly designed to make working with the refiner easier. safetensors. 3. 5 models) to do. set COMMANDLINE_ARGS=--medvram --no-half-vae --opt-sdp-attention. Favors text at the beginning of the prompt. import json from urllib import request, parse import random # this is the ComfyUI api prompt format. . Note that for Invoke AI this step may not be required, as it’s supposed to do the whole process in a single image generation. Lora. 9 comfyui (i would prefere to use a1111) i'm running a rtx 2060 6gb vram laptop and it takes about 6-8m for a 1080x1080 image with 20 base steps & 15 refiner steps edit: im using Olivio's first set up(no upscaler) edit: after the first run i get a 1080x1080 image (including the refining) in Prompt executed in 240. I wanted to share my configuration for ComfyUI, since many of us are using our laptops most of the time. The ONLY issues that I've had with using it was with the. In this guide, we'll set up SDXL v1. 0 A1111 vs ComfyUI 6gb vram, thoughts self. Join me as we embark on a journey to master the ar. Reply reply litekite_ For some workflow examples and see what ComfyUI can do you can check out: ComfyUI Examples Installing ComfyUI Features. Just wait til SDXL-retrained models start arriving. By becoming a member, you'll instantly unlock access to 67 exclusive posts. The workflow I share below is based upon an SDXL using base and refiner models both together to generate the image and then run it through many different custom nodes to showcase the different. Okay, so it's complete test out and refiner is not used as img2img inside ComfyUI. 0 Refiner & The Other SDXL Fp16 Baked VAE. In part 1 ( link ), we implemented the simplest SDXL Base workflow and generated our first images. bat file to the same directory as your ComfyUI installation. SDXL includes a refiner model specialized in denoising low-noise stage images to generate higher-quality images from the base model. ComfyUI also has a mask editor that can be accessed by right clicking an image in the LoadImage node and "Open in MaskEditor". Detailed install instruction can be found here: Link to the readme file on Github. I can run SDXL 1024 on comfyui with an 2070/8GB smoother than I could run 1. SDXL places very heavy emphasis at the beginning of the prompt, so put your main keywords. This was the base for my. Part 4 (this post) - We will install custom nodes and build out workflows. Fully supports SD1. 9 the latest Stable. ago. 5. Reply reply litekite_For some workflow examples and see what ComfyUI can do you can check out: ComfyUI Examples Installing ComfyUI Features. Nodes/graph/flowchart interface to experiment and create complex Stable Diffusion workflows without needing to code anything. download the Comfyroll SDXL Template Workflows. 3. 4/5 of the total steps are done in the base. , this workflow, or any other upcoming tool support for that matter) using the prompt?Is this just a keyword appended to the prompt?You can use any SDXL checkpoint model for the Base and Refiner models. In the case you want to generate an image in 30 steps. Must be the architecture. download the SDXL models. 1 and 0. Thanks. Denoising Refinements: SD-XL 1. They compare the results of Automatic1111 web UI and ComfyUI for SDXL, highlighting the benefits of the former. 这一期呢我们来开个新坑，来讲sd的另一种打开方式，也就是这个节点化comfyUI。那熟悉我们频道的老观众都知道，我一直是用webUI去做演示和讲解的. Examples shown here will also often make use of these helpful sets of nodes: This is the image I created using ComfyUI, utilizing Dream ShaperXL 1. Functions. 20:43 How to use SDXL refiner as the base model. but ill add to that, currently only people with 32gb ram and a 12gb graphics card are going to make anything in a reasonable timeframe if they use the refiner. You can use this workflow in the Impact Pack to regenerate faces with the Face Detailer custom node and SDXL base and refiner models. If you have the SDXL 1. 5. When you define the total number of diffusion steps you want the system to perform, the workflow will automatically allocate a certain number of those steps to each model, according to the refiner_start. It might come handy as reference. SDXLをGoogle Colab上で簡単に使う方法をご紹介します。 Google Colabに既に設定済みのコードを使用することで、簡単にSDXLの環境をつくりあげす。また、ComfyUIも難しい部分は飛ばし、わかりやすさ、応用性を意識した設定済みのworkflowファイルを使用することで、すぐにAIイラストを生成できるように. source_folder_path = '/content/ComfyUI/output' # Replace with the actual path to the folder in th e runtime environment destination_folder_path = f '/content/drive/MyDrive/ {output_folder_name} ' # Replace with the desired destination path in you r Google Drive # Create the destination folder in Google Drive if it doesn't existI wonder if it would be possible to train an unconditional refiner that works on RGB images directly instead of latent images. This workflow uses similar concepts to my iterative, with multi-model image generation consistent with the official approach for SDXL 0. Note: I used a 4x upscaling model which produces a 2048x2048, using a 2x model should get better times, probably with the same effect. 4. sdxl-0. Do I need to download the remaining files pytorch, vae and unet? also is there an online guide for these leaked files or do they install the same like 2. json. . 0已更新！遥遥领先~快来看看更新内容及使用体验~，免费开源AI音乐：文本生成音乐，使用Riffusion实现音乐实时生成，【AI绘画】SDXL进阶篇：如何生成不同艺术风格的优质画面In the realm of artificial intelligence and image synthesis, the Stable Diffusion XL (SDXL) model has gained significant attention for its ability to generate high-quality images from textual descriptions. 下载此workflow的json文件并把他Load加载到comfyUI里，即可以开始你的sdxl模型的comfyUI作图之旅了。 . 0 Base and Refiners models downloaded and saved in the right place, it should work out of the box. png . The result is a hybrid SDXL+SD1. No the 'SDXL refiner' must be separately selected, loaded, and run (In the Img2Img Tab) after the initial output is generated using the SDXL base model in Txt2Img Tab. I can tell you that ComfyUI renders 1024x1024 in SDXL at faster speeds than A1111 does with hiresfix 2x (for SD 1. I was able to find the files online. g. 0 Base SDXL 1. 0 mixture-of-experts pipeline includes both a base model and a refinement model. Works with bare ComfyUI (no custom nodes needed). This is pretty new so there might be better ways to do this, however this works well and we can stack Lora and Lycoris easily, then generate our text prompt at 1024x1024 and allow remacri to double. Edit: Got SDXL working well in ComfyUI now, my workflow wasn't set up correctly at first, deleted folder and unzipped the program again and it started with the correct nodes the second time, don't know how or why. could you kindly give me. SDXL Models 1. I'm creating some cool images with some SD1. Great job, I've tried to using refiner while using the controlnet lora _ canny, but doesn't work for me , only take the first step which in base SDXL. Outputs will not be saved. Here Screenshot . In my experience t2i-adapter_xl_openpose and t2i-adapter_diffusers_xl_openpose work with ComfyUI; however, both support body pose only, and not hand or face keynotes. 9 Model. Create and Run SDXL with SDXL. In addition to that, I have included two different upscaling methods, Ultimate SD Upscaling and Hires. Stability is proud to announce the release of SDXL 1. Omg I love this~ 36. Model loaded in 5. SDXL Refiner 1. 4. When all you need to use this is the files full of encoded text, it's easy to leak. This is a simple preset for using the SDXL base with the SDXL refiner model and correct SDXL text encoders. You may want to also grab the refiner checkpoint. I've been using SDNEXT for months and have had NO PROBLEM. 0 with both the base and refiner checkpoints. A switch to choose between the SDXL Base+Refiner models and the ReVision model A switch to activate or bypass the Detailer, the Upscaler, or both A (simple) visual prompt builder To configure it, start from the orange section called Control Panel. py --xformers. I think this is the best balanced I. I tried comfyUI and it takes about 30s to generate 768*1048 images (i have a RTX2060, 6GB vram). Like many XL users out there, I’m also new to ComfyUI and very much just a beginner in this regard. 色々細かいSDXLによる生成もこんな感じのノードベースで扱えちゃう。 852話さんが生成してくれたAnimateDiffによる動画も興味あるんですが、Automatic1111とのノードの違いなんかの解説も出てきて、これは使わねばという気持ちになってきました。1. 6B parameter refiner model, making it one of the largest open image generators today. 5 method. Step 4: Copy SDXL 0. The other difference is 3xxx series vs. Custom nodes extension for ComfyUI, including a workflow to use SDXL 1. For example: 896x1152 or 1536x640 are good resolutions. To make full use of SDXL, you'll need to load in both models, run the base model starting from an empty latent image, and then run the refiner on the base model's output to improve detail. 1:39 How to download SDXL model files (base and refiner). Custom nodes and workflows for SDXL in ComfyUI. The generation times quoted are for the total batch of 4 images at 1024x1024. ( I am unable to upload the full-sized image. 9, I run into issues. 5 models. SDXL Examples The SDXL base checkpoint can be used like any regular checkpoint in ComfyUI. Table of contents. 5 보다 훨씬 좋아진SDXL을 사용할 수 있게 되었는데훨씬 높아진 퀄리티는 기본에어느 정도의 텍스트 입력도 지원하고그림의 디테일을 보완할 때 사용하는 Refiner도 추가되었다WebUI 에서도 이제 SDXL을 지원하여아래 내용을. 11 Aug, 2023. 5x upscale but I tried 2x and voila, with higher resolution, the smaller hands are fixed a lot better. 手順1：ComfyUIをインストールする. The refiner is although only good at refining noise from an original image still left in creation, and will give you a blurry result if you try. In ComfyUI this can be accomplished with the output of one KSampler node (using SDXL base) leading directly into the input of another KSampler. Img2Img. 0 for ComfyUI, today I want to compare the performance of 4 different open diffusion models in generating photographic content: SDXL 1. Hypernetworks. 9 fine, but when I try to add in the stable-diffusion-xl-refiner-0. 2. For me, it has been tough, but I see the absolute power of the node-based generation (and efficiency). SDXL uses natural language prompts. Play around with different Samplers and different amount of base Steps (30, 60, 90, maybe even higher). ai has released Stable Diffusion XL (SDXL) 1. では生成してみる。. The Tutorial covers:1. SDXL places very heavy emphasis at the beginning of the prompt, so put your main keywords. 5s/it, but the Refiner goes up to 30s/it. Please share your tips, tricks, and workflows for using this software to create your AI art. And I'm running the dev branch with the latest updates. . 手順2：Stable Diffusion XLのモデルをダウンロードする. 5 + SDXL Refiner Workflow : StableDiffusion. {"payload":{"allShortcutsEnabled":false,"fileTree":{"ComfyUI-Experimental/sdxl-reencode":{"items":[{"name":"1pass-sdxl_base_only. I've been trying to find the best settings for our servers and it seems that there are two accepted samplers that are recommended. The SDXL Prompt Styler is a versatile custom node within Comfy UI that streamlines the prompt styling process. separate. Fine-tuned SDXL (or just the SDXL Base) All images are generated just with the SDXL Base model or a fine-tuned SDXL model that requires no Refiner. 0: An improved version over SDXL-refiner-0. Updated ComfyUI Workflow: SDXL (Base+Refiner) + XY Plot + Control-LoRAs + ReVision + ControlNet XL OpenPose + Upscaler . SDXL for A1111 – BASE + Refiner supported!!!! Olivio Sarikas. Table of Content. Pixel Art XL Lora for SDXL -. Working amazing. 0 Resource | Update civitai. 0, with refiner and MultiGPU support. x, SD2. ago GianoBifronte ComfyUI with SDXL (Base+Refiner) + ControlNet XL OpenPose + FaceDefiner (2x) ComfyUI is hard. After gathering some more knowledge about SDXL and ComfyUI, and experimenting few days with both I've ended up with this basic (no upscaling) 2-stage (base + refiner) workflow: It works pretty well for me - I change dimensions, prompts, and samplers parameters, but the flow itself stays as it is. Creating Striking Images on. 論文でも書いてある通り、SDXL は入力として画像の縦横の長さがあるのでこのようなノードになるはずです。 Refiner を入れると以下のようになります。 最後に 最後まで読んでいただきありがとうございました。今回は 流行りの SDXL についてです。 Use SDXL Refiner with old models. json format, but images do the same thing), which ComfyUI supports as it is - you don't even need custom nodes. . 9 base+refiner, my system would freeze, and render times would extend up to 5 minutes for a single render. Download the SD XL to SD 1. safetensors files to the ComfyUI file which is present with name ComfyUI_windows_portable file. 0下载公布,本机部署教学-A1111+comfyui，共用模型，随意切换|SDXL SD1. 9_webui_colab (1024x1024 model) sdxl_v1. For my SDXL model comparison test, I used the same configuration with the same prompts. You really want to follow a guy named Scott Detweiler. In Image folder to caption, enter /workspace/img. This is an answer that someone corrects. The SDXL Discord server has an option to specify a style. json: 🦒 Drive. I think his idea was to implement hires fix using the SDXL Base model. ComfyUI seems to work with the stable-diffusion-xl-base-0. 0 involves an impressive 3. ComfyUI_00001_. SDXL_1 (right click and save as) workflow has the SDXL setup with refiner with best settings. 9_comfyui_colab (1024x1024 model) please use with: refiner_v0. Search for “post processing” and you will find these custom nodes, click on Install and when prompted, close the browser and restart ComfyUI. Switch (image,mask), Switch (latent), Switch (SEGS) - Among multiple inputs, it selects the input designated by the selector and outputs it. SDXL uses natural language prompts. 9 the latest Stable. One of its key features is the ability to replace the {prompt} placeholder in the ‘prompt’ field of these. Base SDXL model will stop at around 80% of completion (Use. r/StableDiffusion. . On my 12GB 3060, A1111 can't generate a single SDXL 1024x1024 image without using RAM for VRAM at some point near the end of generation, even with --medvram set. 如下图refiner model生成的图片质量和细节捕捉要好过base model生成的图片，没有对比就没有伤害！ . 0 base checkpoint; SDXL 1. 5 + SDXL Refiner Workflow but the beauty of this approach is that these models can be combined in any sequence! You could generate image with SD 1. 9 was yielding already. In this guide, we'll set up SDXL v1. 0_comfyui_colab のノートブックが開きます。. 9 Tutorial (better than Midjourney AI)Stability AI recently released SDXL 0. I've been having a blast experimenting with SDXL lately. License: SDXL 0. A dark and stormy night, a lone castle on a hill, and a mysterious figure lurking in the shadows. Then I found CLIPTextEncodeSDXL node in advanced section, because someone in 4chan mentioned they got better result with it. • 3 mo. 0 + LoRA + Refiner With Comfy UI + Google Colab fot FREEExciting news! Introducing Stable Diffusion XL 1. 0. I think this is the best balanced I could find. I'm not sure if it will be helpful to your particular use case because it uses SDXL programmatically and it sounds like you might be using the ComfyUI? Not totally. v1. 0 through an intuitive visual workflow builder. 0 Base and Refiners models downloaded and saved in the right place, it should work out of the box. But the clip refiner is built in for retouches which I didn't need since I was too flabbergasted with the results SDXL 0. 5B parameter base model and a 6. It's a LoRA for noise offset, not quite contrast. The refiner is although only good at refining noise from an original image still left in creation, and will give you a blurry result if you try to add. If you want to open it. . install or update the following custom nodes. ComfyUIインストール 3. Searge-SDXL: EVOLVED v4. r/StableDiffusion. Please don’t use SD 1. web UI（SD. 0 and refiner) I can generate images in 2. Other than that, same rules of thumb apply to AnimateDiff-SDXL as AnimateDiff. Below the image, click on " Send to img2img ". Let me know if this is at all interesting or useful! Final Version 3. But, as I ventured further and tried adding the SDXL refiner into the mix, things. In the Comfyui SDXL workflow example, the refiner is an integral part of the generation process. 15. The second setting flattens it a bit and gives it a more smooth appearance, a bit like an old photo. After an entire weekend reviewing the material, I think (I hope!) I got the implementation right: As the title says, I included ControlNet XL OpenPose and FaceDefiner models. 1 Click Auto Installer Script For ComfyUI (latest) & Manager On RunPod. Generated using a GTX 3080 GPU with 10GB VRAM, 32GB RAM, AMD 5900X CPU For ComfyUI, the workflow was sdxl_refiner_prompt_example. SDXL 專用的 Negative prompt ComfyUI SDXL 1. Aug 2. . What Step. But this only increased the resolution and details a bit since it's a very light pass and doesn't change the overall. This is a comprehensive tutorial on understanding the Basics of ComfyUI for Stable Diffusion. from_pretrained(. I use A1111 (comfyui is installed but I don’t know how to connect advanced stuff yet) and I am not sure how to use the refiner with img2img. If it's the best way to install control net because when I tried manually doing it . Adjust the "boolean_number" field to the. During renders in the official ComfyUI workflow for SDXL 0. 1. It has many extra nodes in order to show comparisons in outputs of different workflows. png","path":"ComfyUI-Experimental. With usable demo interfaces for ComfyUI to use the models (see below)! After test, it is also useful on SDXL-1. Txt2Img is achieved by passing an empty image to the sampler node with maximum denoise. 0 Checkpoint Models beyond the base and refiner stages. So I gave it already, it is in the examples. July 14. It is actually (in my opinion) the best working pixel art Lora you can get for free! Just some faces still have issues. (introduced 11/10/23). The only important thing is that for optimal performance the resolution should be set to 1024x1024 or other resolutions with the same amount of pixels but a different aspect ratio. SDXL Lora + Refiner Workflow. 0. 0. ComfyUIでSDXLを動かす方法まとめ. 0. If you're using ComfyUI you can right click on a Load Image node and select "Open in MaskEditor" to draw an inpanting mask. x for ComfyUI; Table of Content; Version 4. 9-usage This repo is a tutorial intended to help beginners use the new released model, stable-diffusion-xl-0. With SDXL as the base model the sky’s the limit. Create a Load Checkpoint node, in that node select the sd_xl_refiner_0. The only important thing is that for optimal performance the resolution should be set to 1024x1024 or other resolutions with the same amount of pixels but a different aspect ratio. StabilityAI have release Control-LoRA for SDXL which are low-rank parameter fine tuned ControlNet for SDXL which. SDXL you NEED to try! – How to run SDXL in the cloud. do the pull for the latest version. I recommend you do not use the same text encoders as 1. ComfyUI Master Tutorial - Stable Diffusion XL (SDXL) - Install On PC, Google Colab (Free) & RunPod, SDXL LoRA, SDXL InPaintingGenerating a 1024x1024 image in ComfyUI with SDXL + Refiner roughly takes ~10 seconds.