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
Full version of Acorn Is Spinning Flux V1.5 in Q4_1 GGUF format.
FAQ
Comments (27)
why am i getting the following in forge when generating images ?AssertionError: You do not have CLIP state dict!
It may mean you don't have the clip selected in the drop down at the top. https://imgur.com/eI4j1ps
It should look similar to this. The clip, the vae and the text encoder all get selected in the one drop down box.
This model is so crazy good. Especially with Euler ancestral normal and 28 Steps!
Thanks. I much prefer DPM2 with Beta or Normal. Euler usually makes the skin too smooth and more plastic looking for me. (I guess I am not sure which version you are talking about specifically. If it is the De-Distilled, then I think Euler does tend to work the best. With all others I get better results with DPM2) I usually go with 28 steps too if I am using one of the full models rather than an 8 Step.
@Seeker70 I used the full version and read your description. So I tried DPM2. But still, I like Euler ancestral more. I also used a Lora and a Nomos Upscaler. Maybe that makes the difference. Nevertheless it's a really good model.
What the point to make merge just from Loras?
Anyone can take basic model and use exactly what Loaras they need, it is much simpler than download 12Gb and see that it draws nipples everywhere just because author doesnt know how to properly set Loara strength in merge.
The point of merging Loras is that currently that is the best method to improve the Flux model. All trained versions that I have come across are substantially lower quality than mine and even lower quality than the base Flux model in most ways.
Well, nothing prevents you from creating merge yourself. But not everyone knows how to do that.
Apart that, merging LORAs into FLUX is actually often better than using LORAs in workflow, cause for whatever reason, results are not same. This doesnt apply just to FLUX, but to other models as well.
Of course, especially when one needs a lot more than one LORA.
@Mescalamba Yeah, for Loras that affect overall style or other elements that you always want, it is much better to merge them than to use the actual Lora each time. Every extra Lora you use when generating has to be loaded and slows down the generation. Once they are merged into the base model, they no longer do that and you can then add other Loras that you only want for a specific image and it won't bog it down so much. For instance, my Acorn Improved Nipples lora is merged into V1.5. There is never a time when I want the terrible Flux default nipples, so merging it in makes sense rather than just running it for each image gen.
Seeker70 Why not release a Lora too? (more downloads for you without diluting the merged model)
beautifai I do release Lora models too. Just most are character Loras. Any Lora I try to create other than a character is done with the idea of merging it to improve my model. I am by no means a real expert at creating Loras though. I have managed to make a few that helped improve the model. The character models are a lot easier.
beautifai I don't mind more downloads, but I don't really get anything out of it except knowing that people appreciate them. Since there is no legitimate creator program, all my buzz just sits here useless. There is no method to cash it out. The only way I can make actual money is if people go to my YouTube channel and watch a bunch of my music videos or buy my albums on Amazon.
Seeker70 by asking you to make Loras I meant, making a lora of the trained model. So technically you upload both the trained model and its lora for people to download whatever works for them.
beautifai Another user created a Lora from my Acorn Is Spinning Flux model. I tried it out and it seems to work reasonably well, but I just didn't see a lot of reason to use it as opposed to just using the model itself. Here it is if you want to check it out.
https://civitai.com/models/705381/acorn-is-spinning-lora?modelVersionId=789002
I get the following error when trying to use this in a comfyai txt2img workflow:
CLIPTextEncode
'NoneType' object has no attribute 'tokenize'
any idea what I'm doing wrong?
I am not sure. I never use it in Comfy. I think most times I have seen any 'NoneType' errors it is having the wrong clip, text encoder or vae selected for the type of model. Probably someone who regularly used Comfy would be better able to help with that.
Probably nothing, switch ComfyUI to 3.26 .. last 3.27 has more bugs than its healthy. You can do that via ComfyUI manager.
This error happens when a checkpoint doesn't have the VAE and Clip baked in. You'll need to add 'DualClipLoader', 'Load VAE' and 'Load Diffusion Model' nodes.
Actually remembered, if you are using DualClip Loader.. you have to select FLUX in type. Selecting SDXL throws this error.
So what i should use in Vae\text encoder tab in SD Forge?
the ones I use are:
t5xxl_fp8_e4m3fn.safetensors
ViT-L-14-TEXT-detail-improved-hiT-GmP-TE-only-HF.safetensors
ae.safetensors
hi, great work! would you mind to share which training tools are you using
All of my training is done through Loras. Lately I have been using fal.ai and https://tensor.art/ to train my Loras since I don't have adequate PC power to do it locally. Once I train a Lora that achieves a goal, I merge it into my model. For instance, V1.5 was created by merging my Acorn Improved Nipples Lora into the V1.1 model. I have a version V1.69 as well that includes a pussy improvement. I haven't got around to uploading it here yet, but it is up on TensorArt. I am still deciding if I want to upload it here or if I am going to create a better Lora first to give it better functionality. So far, my Loras have been used to eliminate the worst aspects of Flux rather than to make those aspects truly great. Baby steps.
@Seeker70 tensorart.com does not exist
@nutronic You are right. I guess it is https://tensor.art/
can you link your workflow how do you make this checkpoint work? problems with clip encoder and vae
I use Forge UI. It has a drop down menu in the top center to select the clip, text encoder and vae. In Comfy I think there are separate nodes for each of those items. https://imgur.com/a/fOSLZ33
this screenshot shows the clip, text encoder and vae that I use.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.