This my first full LCM model. Its also the first model of the NextGenXL version of Hephaistos. Its trained on a merge of Hephaistos and Colossus Project XL with a ton of my own pictures. So because its trained on a merged model (already with own Loras inside) I call it a hybrid model. Version 2.0 is merged with Osiris Project XL https://tensor.art/models/668037070301698060
!USE DPM++ SDE/Karras or DPM++ 2Sa/Karras!
It doesn't require the LCM Lora! Don't use it with this model!
What is LCM? Basically LCM allows models to operate in much lower cfg and steps.. this way this model is about 6x faster than my normal checkpoints. I've tested in with A1111 and comfy.
There are sadly some downsides to it:
Some sampler aren't working anymore..
This is quite a experimental model to be honest. I still need to do more tests with it.
How to use it?
Here are some settings I used for the showcase. I am working on a proper tutorial but I wanted to share this model with you early. I will add more in the future.
V2.0:
I highly recommend use one of the following samplers: DPM++ SDE Karras, DPM++ SDE, DPM++ 2S a Karras, DPM++ 2S a
3-3,5+ cfg, 5-12+ steps, Clip Skip 2, use a highresfix with 0.2- 0.5 denoise and use this Upscaler https://huggingface.co/FacehugmanIII/4x_foolhardy_Remacri
Euler a:
Euler a needs more steps. I used it with 15 step+ with good results.
Other Sampler are NOT working with the model.
I have tested this woth A1111, a beta workflow for COMFY UI can be found here :
https://civarchive.com/articles/2978
V1.0:
I highly recommend to download version 2.0..
Working settings A1111: ( I have tested)
Euler a, 3cfg, 12 steps
Comfy: I have created some Comfy workflows but its still WIP.
https://civarchive.com/articles/2978
Credits:
Thanks to the LCM Team for making all this possible. Here is their website: https://latent-consistency-models.github.io/
This model is trained on base of a merged model (Colossus Project XL) so I will credit all creators (the stuff I used for Colossus Project XL).
Realistic Stock Photo by PromptSharingSamaritan
https://civarchive.com/models/139565?modelVersionId=154593
NewdawnXL by freek22
https://civarchive.com/models/120741?modelVersionId=163104
Those resources I used to create Hephaistos, Talos and Asmodeus:
FenrisXL by freek22
https://civarchive.com/models/122793/fenrisxl
SDVN6-RealArtXL by StableDiffusionVN
https://civarchive.com/models/118114?modelVersionId=128795
first_nsfw_for_sdxl_v1 by sevenof9247
https://civarchive.com/models/118794/firstnsfwforsdxlv1
Perfect Eyes XL by Deizor
https://civarchive.com/models/118427/perfect-eyes-xl
Finenude by finenude
https://civarchive.com/models/117603?modelVersionId=127928
Hella Babes, a SDXL LoRA for better nudity by vfx_ai
https://civarchive.com/models/127842?modelVersionId=139902
Film Like XL by getjinxed
Description
I managed it that it can now produce high quality pictures starting from 5 steps
IMPORTANT:
I highly recommend use one of the following samplers: DPM++ SDE Karras, DPM++ SDE, DPM++ 2S a Karras, DPM++ 2S a, Euler a (with at least 15 steps)
Other sampler besides Euler a doesn't work properly!!!
3-3,5cfg, 5-12+ steps, Clip Skip 2 use a highresfix with 0.2-0.5 denoise and use this Upscaler https://huggingface.co/FacehugmanIII/4x_foolhardy_Remacri
FAQ
Comments (17)
Sadly I had to close the Civitai generation service for this model because the needed samplers are not available there. I will add this feature when civitai provides those in the future.
I'm using DPM++ SDE Karras, CFG 3.0, CLIP 2, 5-8 steps (tried all in between), and the resulting image is always fuzzy. I tried running the hires fix, but that is too slow on my GPU. Upscaling didn't do a whole lot to clear it up. Thoughts? I see your images are nice and sharp...
What UI are you using A1111 or Comfy? Maybe try DPM++ 2s a or DPM++ SDE (not Karras) Euler a works too but it needs more steps.
Are you using any embeddings or Loras?
@Afroman4peace I am using a couple LoRAs. I'm using SD.Next, currently having a strange bug with DPM++ 2M (no 2s currently), but I could try without the karras scheduler.
@MysticDaedra DPM++ SDE Karras, DPM++ SDE, DPM++ 2S a Karras, DPM++ 2S a and Euler a are the only working Samplers. Euler a works fine but with higher steps. I have tested the model with A1111 and Comfy. Some Loras maybe aren't compatible with it. I found out that a Highresfix with Euler a or DPM++ 2S a are highly recommended. I will post a some pictures with those settings in the gallery.
@MysticDaedra A question.. are you using version 2.0 of this model? It seems that you using version 1.0 Is a old model I wouldn't recommend using anymore.
@Afroman4peace hmm I thought I had downloaded the new version, but I might have selected the old one by accident. I'll check when I get home from work. Thanks!
@Afroman4peace Looks like I was indeed using the old version of the checkpoint, the new one is giving me an error in SD.Next. Is it not compatible with diffusers pipeline?
@MysticDaedra I will look into it after work. I need to install SD.Next first.
@Afroman4peace Turns out SD.Next was not detecting the proper pipeline, probably due to LCM-merged models still being so new. I manually set the pipeline from "automatic" to "SDXL" and it is now working!
Thanks! Quite good and fast model! I like natural looking people with well rendered eyes and detailed skin texture with imperfecions. Most other models generate plastic skin, Hephaistos make them lok very real. Arms and hands are also much better than in standard SDXL. But there are some issues: keywords for weather and lighting are not strong enough, and it tends to render woman with big breasts (even if you enter "smal breasts" or "small tits" in propmt and opposite in negative prompt). I use: DPM++ SDE Karras, 8 steps, CFG 3,5, clip ckip 2, no Hires fix. I generate 768x768, then I upscale best images - text to image from the same seed, 1536x1536, HiresFix with 1st pass size 768x768 and Remacri Upscaler gives 6144x6144 with very fine quality, the result is almost the same as in low resolution source.
Thanks for your comment and your review. I will look into the Breasts "scientifically speaking" I have also a special version of it on Tensor.art It's mixed with my Osiris project. This is might be a tiny bit less realistic but allows to use it online. (You also can download it for free here) I am currently working on a update that brings back some realism from this the 2.0 model.
@Afroman4peace One more issue with breasts. It's very difficult to render hidden nudity. I tried to render nude goddess Venus emerging from the sea and covering breasts with her hands, and gave up after one hour of futile effort. Neither positive nor negative prompts prevented flaunting with nudity.
Hi bro!I followed your tips to set up A1111 web, but the quality of the generated image is still very blurry, cfg-3.5,steps-7,DPM++ SDE,4X_foolhardy, I used XYZ to test the denoisin also did not get the result, this XL_LCM checkpoint is not focusing on the prompt lol? I using your prompt works fine, but adding my prompt makes it worse, hard to fix it!
A friend put me on to your model.
Very nice mate.
Tested a bunch of random stuff with it. Will post some.
Good work. :)
Thanks ;-)
I have ran your model in (HUNDREDS) of ways. It gets the prompt, but it will not focus. Please archive this model. It's useless.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.