OpenArt

<h3>What this workflow does</h3><p>This workflow simply </p><ul><li>loads a model</li><li>allows you to enter positive negative prompt</li><li>allows you to adjust basic configurations like seeds, steps etc</li></ul><p>and generates an image.</p><p><br></p><h3>How to use this workflow</h3><p>Use this workflow only if you are sure the base checkpoint embeds a good quality VAE, otherwise check out another this workflow with VAE - &nbsp;<a href="https://openart.ai/workflows/openart/basic-sd15-workflow-with-vae/XlPv7Qz9U3wOIDn5vSxM" target="_blank">https://openart.ai/workflows/openart/basic-sd15-workflow-with-vae/XlPv7Qz9U3wOIDn5vSxM.</a></p>

Basic SD1.5 Workflow 

<h3>What this workflow does</h3><p>This workflow adds an external VAE on top of the basic text-to-image workflow ( <a href="https://openart.ai/workflows/openart/basic-sd15-workflow/lkOtNJ2UexVd6vK0kYhd" target="_blank">https://openart.ai/workflows/openart/basic-sd15-workflow/lkOtNJ2UexVd6vK0kYhd</a>). The VAE model is responsible of converting the latent image into the pixel space. The process is "lossy" and a good VAE model is essential to get good quality images.</p><h3>How to use this workflow</h3><p>This one workflow uses a default good VAE but there are more. Check out the description on Huggingface or CivitAI if the author suggests a specific VAE.</p>

Basic SD1.5 + VAE Workflow

<h3>What this workflow does</h3><p>This workflow loads an additional LoRA on top of the base model. </p><p><br></p><h3>How to use this workflow</h3><p>** LORAs can be daisy-chained! You can have as many as you want **</p>

Basic LoRA Workflow

<p>This is a basic img2img workflow. You can upload a reference image and a prompt to guide the image generation.</p><p>By default, it generates 4 images based on 1 reference image, but you can bypass or remove the Repeat Latent Batch node to generate just 1 image.</p>

Basic SD 1.5 Img2img Workflow

<h3>What this workflow does</h3><p>This workflow can batch generate images instead of one at at time, by using the Latent From Batch node. It helps get more results at a time.</p><p><br></p><h3>How to use this workflow</h3><p>You can change the batch index to tweak the number of generations.</p>

Batch Generation Workflow

<p>This workflow allows you to skip some of the layers of the CLIP model when generating images. This can be useful for getting more creative results, as the CLIP model can sometimes be too specific in its descriptions.</p>

Clip Skip Workflow

<h3>What this workflow does</h3><p>This workflow applies a second pass &nbsp;of KSampler on top of this basic SD 1.5 workflow with an external VAE ( &nbsp;<a href="https://openart.ai/workflows/openart/basic-sd15-workflow-with-vae/XlPv7Qz9U3wOIDn5vSxM" target="_blank">https://openart.ai/workflows/openart/basic-sd15-workflow-with-vae/XlPv7Qz9U3wOIDn5vSxM</a>). Even at the same resolution (ie: without upscaling), the second pass can increase definition and fix small mistakes.</p><p><br></p><h3>How to use this workflow</h3><p>Increasing the DENOISE adds more details but also changes the original reference more (the first pass).</p><p><br></p>

Basic Second-Pass Workflow

<p>This inpainting workflows allow you to edit a specific part in the image. Right click the image, select the Mask Editor and mask the area that you want to change. The grow mask option is important and needs to be calibrated based on the subject.</p><p>In this example we're applying a second pass with low denoise to increase the details and merge everything together.</p>

Basic Inpainting Workflow

<p>In this workflow, the first half of the workflow just generates an image that will be outpainted later. You can replace the first with an image import node.</p><p>In the second half othe workflow, all you need to do for outpainting is to pad the image with the "Pad Image for Outpainting" node in the direction you wish to add. Note that it's still technically an "inpainting" process, so you need to use an inpainting checkpoint and the "VAE Encode (for inpainting)" node.</p>

Basic Outpainting Workflow

<p>When outpainting it happens to get a noticeable seam line between the two pieces. This workflow applies a low denoise second pass over the outpainted image to fix any glitch.</p>

Outpainting With Seam Fix

<h3>What this workflow does</h3><p>This is essentially the same as the basic workflow for SD 1.5 ( <a href="https://openart.ai/workflows/openart/basic-sd15-workflow/lkOtNJ2UexVd6vK0kYhd" target="_blank">https://openart.ai/workflows/openart/basic-sd15-workflow/lkOtNJ2UexVd6vK0kYhd</a> ), but it's not optimized for SDXL. We include it here for the demo purpose.</p><p>We highly recommend you to use another basic SDXL workflows - &nbsp;<a href="https://openart.ai/workflows/openart/basic-sdxl-workflow/P8VEtDSQGYf4pOugtnvO" target="_blank">https://openart.ai/workflows/openart/basic-sdxl-workflow/P8VEtDSQGYf4pOugtnvO</a> which includes ClipTextEncodeSDXL for better image quality results.</p>

Ultra Basic SDXL Workflow (Not Recommended)

<h3>What this workflow does</h3><p>This basic workflow runs the base SDXL model with some optimization for SDXL. This can be useful for systems with limited resources as the refiner takes another 6GB or ram.</p><h3>How to use this workflow</h3><p>If your model is based on SD 1.5, use this basic workflow instead - &nbsp;<a href="https://openart.ai/workflows/openart/basic-sd15-workflow/lkOtNJ2UexVd6vK0kYhd" target="_blank">https://openart.ai/workflows/openart/basic-sd15-workflow/lkOtNJ2UexVd6vK0kYhd.</a></p><p><br></p>

Basic SDXL Workflow

<h3>What this workflow does</h3><p>This workflow adds a refiner model on topic of the basic SDXL workflow ( <a href="https://openart.ai/workflows/openart/basic-sdxl-workflow/P8VEtDSQGYf4pOugtnvO" target="_blank">https://openart.ai/workflows/openart/basic-sdxl-workflow/P8VEtDSQGYf4pOugtnvO</a> ).</p><p>The core of the composition is created by the base SDXL model and the refiner takes care of the minutiae. The refiner helps improve the quality of the generated image.</p>

Basic SDXL + Refiner Workflow

<p>This is a basic img2img workflow on top of our basic SDXL workflow ( <a href="https://openart.ai/workflows/openart/basic-sdxl-workflow/P8VEtDSQGYf4pOugtnvO" target="_blank">https://openart.ai/workflows/openart/basic-sdxl-workflow/P8VEtDSQGYf4pOugtnvO</a> ). Instead of starting with a random latent image, the workflow will start with a user-uploaded image.</p>

Basic SDXL Img2Img Workflow

<h3>What this workflow does</h3><p>This is the simplest and probably less expensive method to upscale an image on the pixel level. A lanczos algorithm is used in the pixel space to upscale the image and then a second pass applied to it.</p><p>It needs a relatively high denoise which means the final image will be a little different from the source.</p>

Pixel Upscale Workflow

<h3>What this workflow does</h3><p>This workflows loads an existing upscaling model that takes care of the upscaling phase. It's more effective but more expensive than our basic image upscaling workflow ( <a href="https://openart.ai/workflows/openart/basic-image-upscale-workflow/REGMkdFKDcRg4RkpMzvG" target="_blank">https://openart.ai/workflows/openart/basic-image-upscale-workflow/REGMkdFKDcRg4RkpMzvG</a> ).</p><p>The second pass is performed to add back some sharpness to the image but can be done at a very low denoise.</p><p><br></p>

Model-based Pixel Upscale Workflow

<p>The latent can be upscaled directly but the loss in quality is important. That means that the second pass after the upscale has to be done at a very high denoise (0.55+).</p><p>This option is offered as it's very inexpensive computationally and it can be useful if you don't need the final image to be 100% of the initial generation.</p><p>If the final image can be rather different from the first pass, you should use latent upscale as it's faster and very light computationally.</p>

Latent Upscale Workflow

<p>CANNY CONTROLNET</p><p>================</p><p>Canny is a very inexpensive and powerful ControlNet. It extracts the main features from an image and apply them to the generation.</p><p>Remember to play with the CN strength. You generally want to keep it around .7 to give a little leeway to the main checkpoint.</p>

ControlNet Canny

<p>DEPTH CONTROLNET</p><p>================</p><p>If you want to use the "volume" and not the "contour" of a reference image, depth ControlNet is a great option.</p><p>As always with CN, it's always better to lower the strength to give a little freedom to the main checkpoint.</p>

ControlNet Depth

<p>OpenPose ControlNet</p><p>===================</p><p>Basic workflow for OpenPose ControlNet. It extracts the pose from the image.</p><p>It's always a good idea to lower slightly the STRENGTH to give the model a little leeway.</p>

ControlNet OpenPose

<p>DWPOSE Preprocessor</p><p>===================</p><p>The pose (including hands and face) can be estimated with a preprocessor. Probably the best pose preprocessor is DWPose Estimator.</p><p>In this workflow we transfer the pose to a completely different subject.</p>

ControlNet DWPose

<p>Of course it's possible to use multiple controlnets.</p><p>In this example, we're chaining a Depth CN to give the base shape and a Tile controlnet to get back some of the original colors.</p><p>It's important to play with the strength of both CN to reach the desired result.</p>

Multiple ControlNets

<h3>What this workflow does</h3><p>This workflows is a very simple workflow to use IPAdapter IP-Adapter is an effective and lightweight adapter to achieve image prompt capability for stable diffusion models.</p><p><br></p><h3>How to use this workflow</h3><p>The IPAdapter model has to match the CLIP vision encoder and of course the main checkpoint. All SD15 models and all models ending with "vit-h" use the SD15 CLIP vision. One of the SDXL models and all models ending with "vit-g" use the SDXL CLIP vision.</p><p>PLUS models use more tokens and are stronger. LIGHT models have a very light impact. FACE and FULL-FACE are only to describe faces (they are not face swap!)</p>

Basic IPAdapter

<p>It's possible to send multiple images to the same IPAdapter using the Batch Images node.</p>

IPAdapter Batches

<p>FACE MODEL</p><p>==========</p><p>Face models only describe the face. The reference image has to be cut so that only the face is visible.</p><p>Always use square images. Remember to lower the WEIGHT of the IPAdapter.</p>

IPAdapter Face

IPAdapter Inpainting

<p>ATTENTION MASKING</p><p>=================</p><p>IPAdapter supports attention masking.</p><p>The mask determines the area where the IPAdapter will be applied and should have the same size of the final generated image.</p>