If you like my work then drop a 5 review and hit the heart icon
Character commission is open on Patreon
Join my New Discord Server
This model has no VAE baked in so you don't need to use any vae with the baked version.
Suggested settings:
I had CLIP skip 2 on every image
I had ENSD: 31337 all of them
All of them had highres.fix or img2img at higher resolution.
I don't use restore faces
NOTE: if you find any prompt of the preview images look familiar it's because I've taken them from other model review images, Credits to the original authors. Thanks for the benchmark.
Description
FAQ
Comments (26)
Well, the v3 only seems to produce black images. I tried a few different settings and workflows. Following the progress on ksampler node's live preview shows the image to be black on every sampling step in all cases. Looking at CMD window, there's no error messages or anything else.
I grabbed the v2(baked) just to compare, and it works fine.
i tested it on my system and its working just fine. dont know why its producing black image for you
@Goofy_Ai Yeah, I've looked a few more times but no change. I forgot to mention that I'm using ComfyUI, though I don't see how it could have such an effect. My install is also pretty vanilla with only several custom node packs, and I tested with the simplest basic workflow to avoid any influence from those. I'm not good at python, else I would try a little debugging to see what's going on.
@FeatureEnvy can you check on any other ui like automatic1111 or vlad diffusion
@Goofy_Ai Ok, took a bit to get A1111 installed and running. The model works without problems on it. So I guess there's a difference in how ComfyUI interfaces with model files? O_o I don't know other stable diffusion users so can't tell if it is just me having some problem, or Comfy in general.
@Goofy_Ai Don't know if this is any help, but with my very basic undestanding of python and the help of a few youtube vids, I poked at Comfy's nodes. It appears that the clip node is unable to turn text input into tensor values. Where other models return a bunch of unit vectors, this model this returns just a long list of NaNs (Not a Number).
i have problems with v3 too, i dont get black images but weird ones lol, i test it in confyui and A1111
@Goofy_Ai Forge spits out black images too
@Cybrex I confirm this that Forge has some issue with this checkpoint... running on 1.6 has no issue
Can confirm too, only black images. Im using Stable Diffusion Forge
Anyone got a solution to this? I have the same problem both in latest ComfyUI and A1111. Really want to try v3 but I'm not gonna rollback to A1111 1.6
@FeatureEnvy Still no solution?
@Goofy_Ai V2(baked）is wonderful,i use comfyui
Only black images with V3 on Stable Diffusion Forge
Traceback (most recent call last):
File "D:\Programme\Forge Stable Diffusion\webui\modules\extra_networks.py", line 135, in activate
extra_network.activate(p, extra_network_args)
File "D:\Programme\Forge Stable Diffusion\webui\extensions-builtin\Lora\extra_networks_lora.py", line 43, in activate
networks.load_networks(names, te_multipliers, unet_multipliers, dyn_dims)
File "D:\Programme\Forge Stable Diffusion\webui\extensions-builtin\Lora\networks.py", line 58, in load_networks
errors.display(e, f"loading network {network_on_disk.filename}")
AttributeError: 'NoneType' object has no attribute 'filename'
what about automatic1111?
@Goofy_Ai Not sure, but forge and ComfiyUI are very common in use and Forge is based on automatic1111. It should work in best cas for them but I can understad if you are not willing to spend the time fixing it
@Goofy_Ai I have the same issue, ComfyUI just generates black images and a1111 throws an error upon loading the model. 😥 Any idea how to fix this? I tried various settings and even copied the exact prompt/sampler settings from your showcase images and configured things like ESND in the model description but it doesn't work. Tried a separate VAE instead of the baked in one, doesn't work either.
I've installed a fresh version of a1111 without any command line flags, custom settings or extensions to prevent that from interfering and now the model loads, but only generates gibberish like this. Tried a couple of different settings again and also tried your exact settings but to no avail. I'm no expert but I suspect something with the baked in VAE is wonky.
@2shinrei Did you find a solution?
The original autor dont care for it
@Noxturn Nope, I've abandoned the model short after my last comment.
i use comfyui,the same,output image,only black images,whats wrong with me.use another model,dont have this problem.
@Goofy_Ai the same to me,i use comfyui
Still no hope that it get fixed for Forge / ReForge? :(
I see a lot of potential with this but it's lack of ability to work on WebUI Forged is a dealbreaker.