v5 pack -- (install info, show more, scroll down) <div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/uHYhw-2xBiw" start="0"></iframe></div> v5 pack --<div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/Oh1qCxg3u38" start="0"></iframe></div>v4 pack --<div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/MCjGKy3qqe8" start="0"></iframe></div>v3 pack --<div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/3-7FWMO1J4c" start="0"></iframe></div>v2 pack<div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/QbUs9_eYgLM" start="0"></iframe></div>v1 pack<div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/3Vcdg5ocjO4" start="0"></iframe></div><h1 id="koda-collection-ya92n3gwv">KODA Collection</h1>New Kolors Base Model<h3 id="with-ip-adapter-plus-in-comfyuiv1-pack-p83kk9qhg">with IP adapter plus in ComfyUI v1 pack</h3>- base txt2img and img2img workflow - base Kolors IP adapter-plus<h3 id="v2-pack-odw5mkvi4">v2 pack</h3>- Advanced IP Adapter workflow with SUPIR Upscaler- Base workflows for running Hyper Kolors / LCM Kolors<h3 id="v3-pack-ot1q4o86p">v3 pack</h3>- ip adapter embeds- all in one workflow- SUPIR upscaling<h3 id="v4-pack-z09hxyao3">v4 pack</h3><ul><li>12K, 3 stage upscaling</li><li>SDXL checkpoint merge, UNET Block Weighted</li><li>Sampler Garden (Tuned Grid)</li></ul><h3 id="v5-pack-rccvddyqs">v5 pack</h3><ul><li>all-in-one type workflow</li><li>controlnet canny + depth (toggle)</li><li>IP Adapter advanced (toggle)</li><li>Sampler Garden (toggle)</li><li>12K upscaling (2 stage toggle)</li></ul><h3 id="v6-pack-dmg1ywcbg">v6 pack</h3><ul><li>all-in-one type workflow</li><li>controlnet canny + depth (toggle)</li><li>IP Adapter 2 stage (toggle)</li><li>Sampler Garden (toggle)</li><li>Face model creator &amp; swapping toggle</li><li>12K upscaling with color match/remix (2 stage toggle)</li><li>auto prompt enhancer variant workflow</li></ul><h3 id="how-to-use-the-v6-workflow-njf1hz5iz">How to use the V6 Workflow?</h3>The main control panel has ben refactored for easier operation. <img src="https://i.gyazo.com/a47216c2e1a315291041a729edf8c98b.png" />Here we can use the txt2img basic Kolors and get simple output images from prompts When we have placed images into the IP Adapter, we can enable that like so: <img src="https://i.gyazo.com/7e14390471e94a686197f65f71962a0d.png" />As shown above, model switch (2) is used to activate the IPAdapter. You can use the controlnets with or without the IPAdapter, by using that switch. <img src="https://i.gyazo.com/949f45523aebf0036f1e41a1e71fbf7d.png" />When you have finished making some good images, you can activate the Upscaling section:<img src="https://i.gyazo.com/d5dc0e9807a3f2eff4b8d859e9c8ac3f.png" />all dimensions are being driven by AspectSizeV2, fully explained in the video.<img src="https://i.gyazo.com/11df1475062f26bb1c660350155f0fe1.png" />With the new Upscaler, you have two options, Color match and Color Remix:<img src="https://i.gyazo.com/1eaaae4d1e621845db8284661fa5c2d4.png" />You can choose to use the same image for the color match or experiment with other images to alter the color range used. The second new workflow used "Auto-Prompt" Enhancers for both positive and negative. Enter your simple prompt into the left most box, and use the options to enhance your prompt with magic prompt, one button prompt and auto negative.<img src="https://i.gyazo.com/8f8d75a9008aaf38691b3c17c9b9ef42.png" />it is identical to the new all-in-one workflow, just with this added as an example of what you can do quickly with this. It will introduce many obscure tokens and has both style drop down list features and insanity levels for positive and negative. Full previews are included so you can see what was added. <h3 id="how-to-use-the-v5-workflow-agyvxqrc1">How to use the V5 workflow?</h3>txt2img with Cnet+Ipa (Koda-Kolors-Zero-Legacy)<ol><li>Base Kolors modemodel switch (1)Kolors Basic enabled onlyFunctions as only txt2img</li><li>IPadapter modemodel switch (2)Kolors basic &amp; (2) IPadapter enabled</li><li>Controlnet Modecan be used with both model postions, as you choose.enable the Depth &amp; Canny groups together, for default behaviour</li><li>All perturbed modes (sampler garden)these can be use with or without controlnet or IPA modesyou can copy these and customise them to build a dynamic Grid</li><li>12K Upscalingstage 1, 2, 3 each is performing 2X upscale, you can use your best upscaling model toocomparer located at the bottom for easy access, but beware 91mb images can cause UI lag ! recommend to minimise/collapse/remove all previews in stage 2 and 3 ;)</li></ol><h3 id="how-to-use-the-v4-workflow-ly2pypmbq">How to use the V4 workflow?</h3>below pictured: Stage 4 active, switch position 4, [4] combine Embeds.<img src="https://i.gyazo.com/d7d3ff2e88085352dd5b8c7f5895f7ac.png" />the new all-in-one workflow is multifunctional, you use it in stages 1,2,3,4 however it's much easier to use. Simply match the model switch to the active group. All is explained in the v3 video, and in notes in the workflow.Step 1. txt2img onlymodel switch 1used when all groups are bypassedThis is Kolors txt2imgGenerate images and place your best ones into the IP Adapter image load nodesThe rightmost image loader will also be used as the negative embeddingStep 2. a) IP Adaptermodel switch 2requires group [2] IP Adapter, to be activeTest the output of the chosen IPadapter image batch, remember this is separate from the prompt entirely. If you are happy with the results;Step 2. b) Save Embedsmodel switch 2requires group [2] IP Adapter to be activesaves the positive and negative as IPA embed filesthese are saved to \models\output\run once and then copy the files to \models\input\then click refresh in comfyUIStep 3. Load Embedsmodel switch 3requires group [3] Load Embeds, to be activeonce you have created your embed, we should test it.load the positive and negative, then test the generations.repeat the process with characters, styles and scenes so you have some elements to be mixed, then we can head to the final stage.Step 4. Combine Embedsmodel switch 4requires group [4] Combine Embeds, to be activeSelect the embeds and add or remove nodes as neededinterpolation and operators are available, with all the usual IP Adapter nodes.This can work like Aesthetic Embeddings worked with SD1.5For most purposes, we want to anchor in a style or concept and then use text prompts to generate, LORA is the best for this, however this can get very close within the constraints of the model memory.the SUPIR group can be toggled with any stage for easy upscaling~<h3 id="you-will-need:-wn7hvtjnj"> You will need:</h3><ul><li>install any and all missing custom nodes &amp; use the latest version of ComfyUI</li><li>watch the v1 video for more setup info</li></ul> <a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/ChatGLM3-safetensors/tree/main">https://huggingface.co/Kijai/ChatGLM3-safetensors/tree/main</a>download one of the models, i will use the fp8 version, you might use another.place inside ComfyUI\models\LLM\checkpoints\<a target="_blank" rel="ugc" href="https://huggingface.co/Kwai-Kolors/Kolors/resolve/main/unet/diffusion_pytorch_model.fp16.safetensors">https://huggingface.co/Kwai-Kolors/Kolors/resolve/main/unet/diffusion_pytorch_model.fp16.safetensors</a>place inside ComfyUI/models/unet/diffusers/kolors<a target="_blank" rel="ugc" href="https://huggingface.co/Kwai-Kolors/Kolors">https://huggingface.co/Kwai-Kolors/Kolors</a>I also added all of these files from diffusers here. we already downloaded the "diffusion_pytorch_model.fp16.safetensors" and put it into the /models/unet/kolors/ some of the nodes look in a different location, it's up to you which method you want to use as both will work, this is discussed in the video.We do not need any .bin of the contents of text_encoder because we took the ChatGLM3.fp8.safetensors instead. structure of ComfyUI\models\diffusers\Kolors\<pre><code>ComfyUI\models\diffusers\Kolors&gt; tree /F
│ model_index.json
│
├───scheduler
│ scheduler_config.json
│
├───text_encoder
│ config.json
│ tokenizer.model
│ tokenizer_config.json
│ vocab.txt
│
└───unet
 config.json
 diffusion_pytorch_model.fp16.safetensors</code></pre><h3 id="kolors-ip-adapter-gw3sw7mqx">Kolors IP Adapter</h3><a target="_blank" rel="ugc" href="https://huggingface.co/Kwai-Kolors/Kolors-IP-Adapter-Plus">https://huggingface.co/Kwai-Kolors/Kolors-IP-Adapter-Plus</a>download ip_adapter_plus_general.binplace inside \models\ipadapter\kolors\<h3 id="kolors-clip-vision-6vnc4boac">Kolors Clip Vision</h3><a target="_blank" rel="ugc" href="https://huggingface.co/Kwai-Kolors/Kolors-IP-Adapter-Plus/tree/main/image_encoder">https://huggingface.co/Kwai-Kolors/Kolors-IP-Adapter-Plus/tree/main/image_encoder</a>download pytorch_model.binplace inside \models\clip_vision\kolors\<a target="_blank" rel="ugc" href="https://civarchive.com/models/441432/mistoline">https://civarchive.com/models/441432/mistoline</a>place inside \models\controlnet\kolors\ I'm suggesting a subfolder, or if you prefer rename them as needed.<h3 id="vae-5x4rthpmq">VAE</h3>For all the VAE - just use your favorite SDXL VAE<h3 id="controlnet-d7w02o86i">CONTROLNET</h3><a target="_blank" rel="ugc" href="https://huggingface.co/Kwai-Kolors/Kolors-ControlNet-Depth/resolve/main/diffusion_pytorch_model.safetensors?download=true">https://huggingface.co/Kwai-Kolors/Kolors-ControlNet-Depth/resolve/main/diffusion_pytorch_model.safetensors?download=true</a>download and rename to kolors_controlnet_depth.safetensors<a target="_blank" rel="ugc" href="https://huggingface.co/Kwai-Kolors/Kolors-ControlNet-">https://huggingface.co/Kwai-Kolors/Kolors-ControlNet-</a><a target="_blank" rel="ugc" href="https://huggingface.co/Kwai-Kolors/Kolors-ControlNet-Canny/resolve/main/diffusion_pytorch_model.safetensors?download=true">Canny/resolve/main/diffusion_pytorch_model.safetensors?download=true</a>download and rename to kolors_controlnet_canny.safetensors and place both inside \models\controlnet\kolors\ ~Would not be possible without the efforts of@cubiq [IP Adapter nodes]@kijai [Kolors Wrapper]@MinusZoneAI [Kolors Ipadapter nodes]Cubiq<a target="_blank" rel="ugc" href="https://github.com/cubiq/ComfyUI_InstantID">https://github.com/cubiq/ComfyUI_InstantID</a>Kijai<a target="_blank" rel="ugc" href="https://github.com/kijai">https://github.com/kijai</a>Minus Zone AI<a target="_blank" rel="ugc" href="https://github.com/MinusZoneAI">https://github.com/MinusZoneAI</a>

Koda KOLORS Pack - Controlnets,12K Upscale, IPAdapter in ComfyUI

Kolors is a base model, SUPIR used SDXL Hyper but is optional.

Files with SHA256: c81a7f74890c9cf4783c718db0775d3ddaf3f435568d813956d2b06dc5405ba8