Pony/Illustrious models are cool. SDXL is cool. But each has its own limitations. What if we could combine their strengths, taking the best of both worlds? With this workflow, you can produce really believable pictures of unrealistic concepts.
COYOTTE'S REFINER
That workflow used to be a full article for a really long time (7 months). But it's just now I've realised that this whole time it WASN'T WORKING THE WAY IT SHOULD, thanks @v33tay for pointing me to the righ direction. In fact, he helped to develop the OG version, so be sure to show him some appreciation. By default the workflow runs in the new, proper mode. But I've decided to also leave that "wrong way", so if you want to use it - switch "LEGACY REFINE" node to "true". Sometimes one gives better results than the other.
This update also includes some improvements to make the workflow easier to use, so it was long overdue.
Pony-based and Illustrious-based checkpoints are known for the excellent prompt understanding and good NSFW capabilities. These models, however, often struggle with realism. Even those Pony/Illustrious models that are finetuned to be realistic, still tend to have a cartoonish aesthetic. This workflow utilizes an advanced refining method to enhance Pony/Illustrious models output with SDXL models, so that we can get the best of both worlds - highly coherent SFW/NSFW art with the realism of the best SDXL models.
In this workflow, Pony/Illustrious models serve as the foundational layer, taking advantage of their strong NSFW capabilities, while SDXL models are applied slightly later in the diffusion process to enhance the realism.
I strongly recomend you to use my checkpoint LUSTIFY! Endgame or OLT as the refiner, because both this workflow AND the checkpoint were designed around each other. Also, LUSTIFY! is just really good at this point, if I can say so about my own creation.
GUIDE:
1) One of the key elements of this workflow is "Seed Base" and "Seed Refine". By changing "Seed Base" you're changing the overall composition. Got a composition you like, but the details from SDXL checkpoint are kinda 'meh'? Change the "Seed Refine" - the composition stays the same while the details are completely new.
Another important aspect is the "Refine At Steps" setting. A good range to start with is 10-13. The later the refinement begins, the more influence the Pony checkpoint will have. Feel free to experiment with this to see what works best for your needs.
2) You can find the highres fix, tiled upscale and facedetailer nodes lower in the workflow window. By default they're bypassed. You should unbypass them when you've created a pic you like in the first part.
And the good thing is that you have complete control over these parts as well. Don't like the way highres fix looks? Change the seed and/or denoising! Don't like the face "facedetailer" inpainted? Change the seed and/or denoising.
Description
FAQ
Comments (31)
а нода легаси рефайн откуда берется, у меня красная(
Fixing that bug in your original workflow myself while learning Comfy felt like graduating for me, appreciate the update.
Any chance for a Automatic1111 and Forge tutorial?
I'm being unable to import ComfyUI-GlifNodes
Does anyone knows how to fix it, please?
PLEASE REDOWNLOAD THE WORKFLOW.
I've just replaced it with a version that fixes one small thing that was wrong and could result in the workflow not working at all
Hello Coyotte! Great work on your models, you've been a real inspiration. Used the previous version of your workflow. I'm trying to get the new one working and am running into an issue where loading your workflow somehow breaks comfy for me. I can't click to drag the screen and I can't even load other workflows. I've tried the latest version of your workflow per your other comment. Is there a preferred version of comfyui recommended for this workflow?
Why are the generated pics very noisy when legacy refine is turned off?
Hey when I load this workflow into comfy, it just freezes everything, cannot zoom or drag the workflow around or edit anything. Tested another workflow from someone else and that works fine so do you have any idea what may be causing this?
may i know what function is legacy refine? should i on or off it?
So even after installing Crystools (1.22.1) and restarting Comfy, the workflowLEGACY REFINE node is still missing. Any idea how to fix this?
I'm apparently missing a custom node called "ImagesToMultipleOf". I've tried using ComfyUI manager to download any missing nodes and I've also tried cloning and reconnecting the node. Anyone know if there's a fix or a github link I can clone into my custom nodes?
love this workflow! what loras would be recommended? theres two areas, loras for base and loras for refine
it has big problems with my lora, do you have any idea how to fix this?
I'm using an RTX 5070ti (16GB VRAM). When I generate images with this workflow on my PC, the upscaling step consumes too many resources and slows down my computer. If I disable upscaling in the node options, image generation doesn't work. Please let me know if there's a solution.
Could you elaborate more on seeds, the example image has different seed values on each step. Do you recommend keeping one seed value for the entire workflow or do you change it on each step? your stuff is my fav so i'm trying to learn from it
I have the latest version of ComfyUI, everything is updated, the workflows themselves are good, but, author, how do you plan to fix the bug with “ImagesToMultipleOf”?
Installing ComfyUI-GlifNodes manually and via ComfyUI Manager doesn't help in any way, no other versions of ComfyUI-GlifNodes are available.
In the current form, without intervention, these workflows do not work.
This is great--after some updating nodes and experimenting, finally getting all/most of the aspects to function, and even when I couldn't use the highres fix and face detailer, it still was great. I've found that my favorite combo (thus far) is actually not pairing it with a Pony model, but rather Dreamy Hentai SDXL. With of course the best SDXL checkpoint, Lustify.
btw. the "legacy refine" comes handy if you use different schedulers as "the correct" way just creates a colorful mess whilst the legacy (return with leftover noise = off) continues to render correctly.
Does Forge refiner worth something? Only option to change, when to switch to refiner model. I got only underwhelming results.
It's a bit tricky, but it's a very powerful solution. It can realistically capture the extreme expressions of IL and Pony. The results are slightly noisy, so some adjustments are necessary, but it achieves results without adding numerous Loras in one go. Thank you!
nice flow! how do you command prompt adherence, cfg value isn't there before the upscale?
edit
I think I found out, the float node...
The workflows is great. But I struggled trying to adapt it to img2img. It would be great if you released a proper img2img adaptation...
I tried to install all missing nodes, but "workflowLEGACY REFINE" keeps missing. What is this node and where can I get it?
How do you do a imgximg here? It's just prompt to image
I think we need to organize a pintervention for you. The first stage is accepting you have a problem. No judgement. Also, the P key does exist, so I don't really have a great argument.
I am having a lot of trouble adding ControlNet nodes to control the installation.
Do you recommend your new LUSTIFY checkpoint! [SDXL NSFW checkpoint] GGWP (V7) for this workflow or not?
Please, is the checkpoint for “Pony/Illustrious” the following: CyberRealistic CyberIllustrious V1.0 by Cyberdelia?
Hello, I don't understand how to maintain consistency every time. Please help me.
IS this any different that the refiner in A1111?
Happy New Year! I tried your workflow. All nodes and models are loaded and the final preview is generated, but upscaling and refining aren't running (the last three image nodes are highlighted with a red border).
I noticed that the link from the "UltralysticsDetectorP" node to FaceDetailer's bbox_detector is crossed out with a red cross.
Could you tell me what the problem is?
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.