Please consider buying my two albums Sisters of the Wicked and Danger Ahead, my book Dracula and also please check out my YouTube channel. Civitai won't support creators, so any purchase is a real help. Thanks.
DRACULA https://www.amazon.com/dp/B0FM4DXYPL
Sisters of the Wicked https://amazon.com/music/player/albums/B0DSJY3ZHY?marketplaceId=ATVPDKIKX0DER&musicTerritory=US&ref=dm_sh_CdyeuxIBNBDOjZ1hpB1PPJxT0
Danger Ahead https://www.amazon.com/music/player/albums/B0DB2DV354
Music In The Machine https://www.youtube.com/@musicinthemachine/videos
Lora Creation: https://www.upwork.com/services/product/design-flux-lora-for-creation-of-ai-art-1909718528813429477?ref=project_share
For the De-Distilled Model, Distilled CFG Scale should be set to 0. Then use actual CFG. For me 3.5 to 8 seem good.
If you like the models, please follow. The page is getting crowded as I keep releasing my model in as many formats and options as I can.
V1.1 Hyper 8-Step is now available and works in ForgeUI as well as Comfy. Schnell V1 Much faster, getting good results with 1-4 steps. Quality is not quite on the level of the Dev model, but much faster. So far it seems text is more consistently good though. New GGUF version of the Schnell model up now. Thanks to pkmngotrnr for converting them for me. Much appreciated. So far works in Comfy only at least for me. DEVfp8V1.1GGUF is now up as well. V1 version has European Flux mixed in at what seemed like a pretty low percentage, but it seems like it is a bit too strong. If you like this look, feel free to use this one.
This is my initial FLUX model. It is base on the flux1-dev-fp8 model. This is a merge only in the sense that I merged in some Loras that I like to make it more realistic. As with all of my models, the goal is the best photo realism that I can achieve. If it will do other things well, that is a bonus. Until I can learn a lot more about FLUX, and see if at some point true merges are possible, this is a pretty good start. I don't know about compatibility with all software. I use ForgeUI and I am able to run it fine. Most gens take about a minute and a half for me at 1024 X 1328. I use Euler with the Beta scheduler or the Simple. I haven't experimented much yet with other samplers. I typically use 24 to 28 steps.
Description
This is the Acorn Is Spinning Flux V11 Hyper 8 Step in GGUF Q5_K_M
Works in Forge and Comfy, haven't tried it in any other generators.
FAQ
Comments (41)
This is a great checkpoint, but I'm have a recurring issue with nipples on nude females not rendering right. They almost always come out weird. Is this a known issue? I'm using the 1.1 FP8 version.
This is a base Flux issue. Mine has some improvement over the base Flux, but it hasn't completely overcome the original censorship of Flux. Still working toward it.
@Seeker70 I figured it might be. Wouldn't it be possible to create a good-quality lora with detail of breasts and nipples and merge it into the model? I thought that was possible. I've done some model merging with SDXL and 1.5 to create some custom checkpoints, but I've not attempted to merge a lora into a checkpoint.
@Agkistrodon It is probably possible, but it seems so far, most of the Loras created to do that have been created on SDXL content rather than real photos. Also, in my experience so far with trying to use those, if you mix it in strong enough to correct for the bad original, you also end up with nipples on the outside of clothing, so it only ends up being good for nude shots only which is a pretty low percentage of the overall images I like to create.
@Seeker70 It might be a waste of time with SD3.5 being released lately - there are already 8-12GB fp8 and fp16 versions of it out already, and as far as I know, I don't think it has the built-in censorship that Flux does. I'm hoping that some NSFW models are out within the next month or so, and I'd wager there will be.
@Agkistrodon It's a possibility. I haven't tried SD3.5 yet. Until it is able to run in Forge, I don't really care about it. From what little I have seen of it so far, it seems that it's ability to produce photo real images seems lesser than Flux. Again, haven't used it myself though. I don't like using Comfy.
Bro where is "Boning" ? Thanks !
You might want to remind people on the model card that on de-destilled models they need to use regular cfg and not destilled cfg or else they'll just end up with blurry garbage.
I did put it in the version notes, but maybe I will add it to the main description area as well.
yesss. A de-distilled model fine tune!
Will there be a fp16 release?
I will have to look into it. I am not positive if my PC is able to even work with the full size model. I am out of town visiting family, so it likely would be at least a week until I can work on it.
On my 24gb vram card, I find that I have to use the card solely for generating fp16 images and offload the monitor to the integrated video on the motherboard.
@Seeker70 Yes, please, make a fp16 Distilled Version !
@uprz23736 Very good Idea ! Thanx a lot
There may be Flux finetunes that offer better photorealism but this one beats them all when it comes to prompt adherence. There is no other model available for offline use that follows prompts like AiS does. Flux-dev was already quite good at it but AiS no doubt gets the cake!
@seeker969 Do you happen to have a comfy workflow example for Spinning FLUX?
You can use my latest works for it
Here is a basic workflow I use that includes loaders for the different types (GGUF, NF4, regular).
https://civitai.com/models/897505?modelVersionId=1004289
Flux Butt Chin!!! Can you merge an anti-flux chin type LORA with this please? AntiFlux - Chin, Chin Fixer 2000, FLUX -– Polyhedron: Cleft Chin and Bunny Teeth Fix (with open and closed mouth) [female] AND/OR DaChinFix. Otherwise, a great realistic checkpoint you have here!
I am taking a look at it and doing some testing.
New version out incorporating a Chin Fix. Tried tests with several, but with the Chin Fixer 2000 tests, it seemed to really affect other aspects of the model including worse text performance, so I went with DaChinFix as it seemed to have the least negative affects.
Looking good, I see that the chins have been fixed, no butts to see here, except in the rear where they belong :) Sorry for putting you through the trouble, I wish Black Forest Labs had noticed it before their release.
@afterate2 huh, I just took the new model that supposed to fix this, 5/5 generations have butt chins. Albeit not as bad.
@singularity42 That's strange, I saw your sample images and liked them. It's true that some chins are square and have a little dimple, but much less than the original model, which is 100% better to me. Flux butt chins are way too distracting.
Biggest issue I am having is age of people. Describing a 50 year old person and a 20 year old person means I get two generic 30ish year old people.
Still have the issue where you can't describe a body type, especially every woman having large breasts.
flux is not trained on ages in years ... only young mature old
@sevenof9247 Thanks! i feel dumb not knowing that. Is it only those three?
@AcadiaAugust178 its all experiences ;)
what do you mean with three ?
@sevenof9247 young mature old. Does it understand any other words from it's training to be more specific?
@AcadiaAugust178 , infant, child, very young, wrinkles, folds ...
lets say depend i bit on model ... if some one trained on top only "woman" or "1girl" its a bad trainig
very confusing
what you doo with your models ???
What you have merge ?
What you have trained ?
what is different ... post an image post to compare to FLUX_DEV original
I started with Dev1 FP8 model and merge in a number of Loras. No training. Just a recipe.
@Seeker70 Great recipe ! thanks for this delicious meal ;-)
can de-distilled model work with normal flux loras??
Yes!
Thanks for an update here. Can you confirm that the latest Schnell is fully Apache 2 and you didn't integrate Dev into it?I intend to use it for commercial purposes.
The Schnell model is merged with several Loras only. No merging of checkpoints in any way.
if you'll pardon the impertinence of my question, is the license agreement on the honor system or is there some kind of indelible watermark that the non-free flux version embeds in the image?
@fablegenius Truthfully, I don't really know. The license attaches automatically with the model. I don't ever manually add the license info to it.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.