If you've found value in the model, please consider offering your support. Your contribution will be devoted entirely to advancing the SDXL community.
šš¼āāļø join us (discord) ć ¤|ć ¤ š purchaseć ¤ |ć ¤ š± donate
AlbedoBase XL (SFW&NSFW)
The refiner is unnecessary, and VAE is included.
GOAL
Stable Diffusion XL has 3.5 billion parameters (excluding the Refiner), which is about 3.6 times more than the SD v1.5 version. I believe that this is not just a number, but a number that can lead to a significant improvement in performance.
It has been a while since we realized that the overall performance of SD v1.5 has improved beyond imagination thanks to the explosive contributions of our community. Therefore, I am working on completing this AlbedoBase XL model in order to optimally reproduce the performance improvement that occurred in v1.5 in this XL version as well.
My goal is to directly test the performance of all Checkpoints and LoRAs that are publicly uploaded to Civitai, and merge only the resources that are judged to be optimal after passing through several filters. This will surpass the performance of image-generating AI of companies such as Midjourney.
As of now, AlbedoBase XL v3.1 Large has merged about 200 selected checkpoints and 251 LoRAs.
LOG
v3.1-Large
ā¢ Merged over 50 selected latest versions of SDXL models using the recursive script employed in V3.
The spec grid(370.7 MB): download
v3-mini
I sincerely apologize for keeping you waiting for such a long time.
Iāve been dealing with some personal matters, and while working on the new version, I also faced health issues. Even as I write this, I am still struggling with these challenges.
I felt it wouldnāt be enough to just provide a brief update, so I kindly ask for your understanding as I share this more detailed message.
Since the release of version 2.0, I have been dedicating myself to studying deep learning independently. I donāt have a formal degree, and apart from a modest aptitude for programming, only my background is in the arts. As a result, I lack the mathematical and scientific foundation to achieve significant breakthroughs, given the time and effort Iāve invested. Despite this, the experience of immersing myself in this self-directed study and research has been an invaluable treasure in my life.
Recently, I stumbled upon an idea that could potentially be a significant breakthrough. After reworking hundreds of formulas and methods since version 2.0, I managed to develop a rather intriguing and successful algorithm. The model merging process was based on SDXL1.0 and SD1.5, along with other carefully curated models. These were categorized into five primary classifications: āANIME,ā āREALISM,ā āARTISTIC,ā āNSFW,ā and āBASE,ā and fed into the merging algorithm as datasets. This approach has resulted in some fascinating outcomes.
However, as challenging as the algorithm development was, nothing has been as daunting as the performance testing phase. My physical and mental health deteriorated significantly during this period, to the point where I realized I couldnāt continue this work on my own. This is what ultimately led me to decide to release this version.
And now, I am thrilled to announce the release of the much-anticipated AlbedoBaseXL V3 Mini version. While this model is a smaller-scale merge, it is not limited to any specific area and performs remarkably well across various domains. It has the potential to serve as a new base model for SDXL1.0. (For reference, my merging algorithm is not a ālinear merging,ā so it can essentially be considered a new fine-tuned model.)
This model, along with the existing AlbedoBase models, is versatile and surpasses all previous versions in every aspect. (The NSFW content, while not extreme, offers a broader range of expression compared to previous versions like v2.1. A dedicated NSFW merging model will be released in the future.)
On another note, Iāve noticed that many shared models have recently begun to adopt licenses that prohibit merging or external commercialization. This has been a disappointment, as it has prevented me from using some truly excellent models for merging.
I would like to express my heartfelt gratitude to the model developers who have provided free licenses, allowing their high-quality modelsāproducts of considerable time and effortāto be used for merging.
Iāll be back soon.
I eagerly look forward to your performance tests across a wide range of areas, including ANIME, REALISM, ARTISTIC, 2.5D, 3D, and NSFW.
As model developers, we only plant the seeds. It is ultimately you, the model users and artists, who cultivate them and bring forth the flowers and fruits.
Thank you, as always.
For those who wish to support my work with a small financial contribution, please consider using the links below. I am currently unable to secure employment and face an uncertain future regarding my livelihood.
The spec grid(380.5 MB): download
v2.1
Re-merge and adjust v0.1 to 2.0 using new merging algorithm and formula.
The spec grid(424.5 MB): download
v2.0
I'd like to thank everyone who helped me on the AlbedoBase XL Pre side. Without you guys, the release date would probably have been much later. Thank you so much!
I have written a custom script to converge the existing AlbedoBase XL models into one. Intricately aligning the row and column weights of all U-NET and CLIP blocks according to a unique formula of mine.
If you encounter a bug in image generation (if nothing is generated), please switch to CLIP SKIP 2 or modify the prompt slightly! There may be combinations of prompts that CLIP does not recognize. In that case, you can change the order of words, use different words, or, most simply, change the CLIP SKIP. I will gradually work on resolving these issues in the future like v1.3.
The spec grid(403.5 MB): download
v1.3
In order to illustrate the quality associated with the model's randomness, I standardized the seed value at '9' for all showcase images intended for sampling and proceeded with their immediate generation.
Especially with this version, due to the significant impact of negative prompts, leaving the negative prompt field empty is likely to produce the nice quality.
The spec grid(438.7 MB): download
As you can see, as the number of Steps increases, it becomes available for all samplers, and the quality also improves.
Due to the effect of the LoRA I developed and merged, as described below, using sentence-form prompts rather than tag (a list of words) prompts is directly related to improving quality.
I merged 45 checkpoints and 7 LoRAs. After that, I merged AlbedoBase v0.4 and v0.3 in order, less than 0~5%, to reawaken the diluted merged models that had become outdated.Ā
Among the 7 LoRAs, one is created by me. It involves analyzing and annotating captions for a total of 174 high-quality pictorial photos using GPT4-V. Merging this LoRA resulted in astonishingly clear images and an impressively excellent understanding of prompts.
My self-created LoRAs are exclusively available for purchase to my Ko-fi supporters at the Creative level or higher.
v1.2
Merged the 22 latest checkpoints.
The spec grid(565.6 MB): download
v1.1
Stabilized.
More detailed.
If you think you are an advanced user, I recommend version 1.0. If version 1.0 finds the right settings, it can output much more vivid works.
The spec grid(349.7 MB): download
v1.0
Merged 106 LoRAs.
Merged 19 Checkpoints.
The model can produce different results depending on the settings you choose, so it's important to check the spec grid before using it.
I have found that using a few specific negative prompts can help to resolve the issue of asymmetrical eyes or pixelated images. The Spec Grid may vary depending on your CPU or GPU device, so please use it as a general reference. Experiment with a few negative prompts to improve the quality(ex; strabismus). I found that it is difficult to satisfy all settings equally as the number of merged LoRA increases. However, I would like you to focus on this advantage in version 1.0, as it can produce works of amazing quality in various aspects with the right settings. I will be back with a more stable version in the future.
You can find useful setting values in the showcase or by searching for others.
As always, it's best to leave the negative prompt blank for the best results.
This v1.0 was a lot of work, so I'm taking a break for a bit. I hope you'll enjoy using the model, and if you merge it, please share it in Civitai for free. That way, we can all keep improving it.
The spec grid(479.4 MB): download
v0.4
Merged 132 LoRAs.
Merged 4 Checkpoints.
The spec grid: download
v0.3
Improved in all of the samplers.
Achieved lifelike realism.
Stabilized.
The spec grid: download
v0.2
Significant improvements in clarity and detailing.
Improved hand and foot implementation.
Major aesthetic improvements; composition, abstraction, flow, light and color, etc.
v0.1
After appropriate fine-tuning on the SDXL1.0 model, meticulously and purposefully merge over 40+ high-quality models publicly available on Civitai.
Testing has primarily focused on ensuring maximum quality with the minimum number of prompt tokens, and it has not been confirmed how much the quality can improve when using a large number of tokens. (Please conduct your own tests and share the results)
Typically, the most beautiful results are achieved at the midpoint between reality and animation.
Nevertheless, when using an appropriate prompt, there is generally nothing it cannot express. (I assert that it possesses abundant value as a foundational model that surpasses others in merging. However, please keep in mind that this is currently v0.1)
Description
ā¢ Merged over 50 selected latest versions of SDXL models using the recursive script employed in V3.
FAQ
Comments (44)
I love your work - Albedo produces results that take your breath away in many different styles - realistic to cartoon. Thanks for all your hard work!
Stunning model, lives up to your aspirations for advancing SDXL to new heights. The variety of the images it produces is truly amazing. Very grateful for your extreme effort in creating all your models. Best wishes to you.
low step version is needed.
It can be produced with fewer steps too. Or did you mean the Turbo or Lightning version?
@albedobondĀ 8 STEP
@jeriffĀ In the gallery below, Daemona is using Lightning LoRA with 10 steps, but 8 steps should work as well. Using LoRA should make this quite easy for you
thanks for your work!
I have been using version 2.1 of your model for a long time and I made a merge based on it for my own purposes (for example, I selected merge parameters so that black-and-white photos are not gray - but more like good b&w photos).
Also managed to improve anatomy thanks to the merge with porn model The Araminta Experiment Ev4.
I would very much like to get a similar model for SD 1.5.
Hey! What you done with AlbedoBase XL?!
It's become even better!
New Large model doing fantastic job with my Loras. Thanks!
Thanks! :) Stop by on Discord whenever you have time!
I love your model, it draws very beautifully and clearly :)
Now it's time for the real hardcore NSFW model. If you'd like to test the model preview, please join the Discord below!
https://discord.gg/ky6XF845aF
Combining two of my favorite models gave a great result!
theAramintaExperiment_ev4 + albedobaseXL_v31Large - theAramintaExperiment_ev4 x 0.5_trainDifference
Got a great all-round model that generates perfectly, knows and can do a lot of things including my favorite Soviet theme .
My old LORA's started working much better in the model after this merge!
Thank you for the very valuable information! Seeing the result images as well would be very helpful for the next version. If you're able, would you be willing to join our lab's Discord(https://discord.gg/ky6XF845aF) server and share the results there?
I'm in Russia - Discord doesn't work for us (I don't use VPN, they are blocked too and need to be changed regularly).
The Araminta Experiment model does improve generation, including fingers and toes, and not only people, but the quality of generation in general. But it needs to be measured carefully - it tends to undress people.
I use The Araminta Experiment Ev4 - it seems to me the most adequate.
I trained my Loras on albedobaseXL_v21 - on this merge it started working orders of magnitude better, I am even shocked by the results.
Examples can be seen on my channel on telegram
Exactly my thoughts, my LoRA's with Large model works fantastically Good,
even fixing my own mistakes and problem in my LoRA's.
@cr_cat387Ā Thank you, I understand. I'll merge it and run some tests! By the way, the quality of your work is absolutely insane. You have incredible talent. It's very much my taste as well...
@albedobondĀ Thanks!!! But a lot of credit here goes to the author of The Araminta Experiment model - she really takes both the quality of the generations and the quality of the anatomy to a new level. And it works on mergers with her!
@cr_cat387 Yes, but what you've imagined is entirely yours. I really love the world you're trying to express and the techniques you use to express itself :))) They're just perfectly my taste.
Could you tell me which models are included in the latest model?
In albedobaseXL_v31Large there appeared Soviet symbols in the form of sickle and hammer.
In the new model theAramintaExperiment_Gv1 the same. Apparently in both models added some one and the same model with Soviet symbols - here is its name and I would like to know?
The V3-Large model incorporates approximately 40 models, including those used in the previous V3-mini version. I employed scripts to merge them, both in their entirety and partially, and then further merged those results iteratively. Unfortunately, I lost the model list during this process. I've since embarked on new merging projects and, for disk space management, deleted most of the original models a while ago. Furthermore, these models weren't all downloaded within a specific timeframe, rendering my Civitai download history unhelpful. This is why I simply stated in the changelog that around 50 models were merged, without providing a specific list.
This is a stunning model.
Brilliant and superb.
Works perfectly when enhancing and resampling low-res photos, as well as a base model for training real character LoRAs.
Skin texture, eye details, lips, hair, everything is just brilliant, no match.
Big Thumbs Up!!!
I'm loving this model, thanks so much for your hard work. I am so looking forward to future updates :)
The Albedo Model is fantastic. My Pictures calculate with it are great and detailed. I use this Model for various Themes. š
Am I the only one who still prefers v2.1 vs v3-Large?
agree with 2.1
My go-to model when I need to produce crystal-clear, refined/elegant output. Featured as the base-model in my workflow and Comfy node-pack. Surpasses FLUX in many ways.
Wow. The 3.1 Large model is quite impressive. Very good lighting and diffusion.
Looks amazing. Do you have a recommended set of base stats to start off with?
Been playing with this lately. Leading with prompts that I stole from one of the sample art gets me good results:
(in ARTIST style:1.4), (masterpiece:1.7),
(masterpiece:1.8), (highest detailed:1.8), (angle:1.8), perfectly lighted, (contrast), (saturated), (((sharp focus))),
(hyperdetailed:1.3), (intricate details:1.2), (Refined details:1.3), (best quality:1.5), highly detailed transparent textures, transparent vibrant colors,
<lora:add-detail-xl:1> (ultra sharp focus:1.8), (thin lines:1.8), (best quality:1.8),
(Your prompt here, with a lora or two)
(loose the line about transparent textures and colors as appropriate.
"A dedicated NSFW merging model will be released in the future."
If you do this, could you please merge in a gay model or 2, in order to balance out the generations a bit? I understand if you're straight and don't want to look at that stuff, but, if you're just merging then you don't really gotta look at it.
Doing so would not only do better at creating that content for those that want it, but also would create a more balanced output that would not lean so heavily into female generations to the extent that it actively turns guys into women.
Cuz most of the models are very heavily loaded down with female datasets. I'm not a misogynist or anything of the such, but I do believe in gender equality, which really hasn't been a thing with AI so far.
Or even reduce the amount of females in the data if that's possible, so that the word "beautiful" can be used in a generation to create amazing picturesk landscapes and fantasy images, without that word defaulting to generating half naked women.
Disclaimer: I have not yet used your models, so any of the above is merely based on my experience(with many other models), plus the look of the model names that you've merged thus far. I cannot yet speak for the biases in your models. Though I do plan on trying out the newest version after I get some sleep.
Hey did you find any?
@BoyWithToysĀ If you mean 'gay models', the best I've found so far, is https://civitai.com/models/964011?modelVersionId=2449414
I typically use 2.0 instead of 3.0, as, iirc, it generates better fem boys out of the two (it makes sense, as, more men trained in, = more muscle), Though, I'm a bit outta date on them, I haven't even tried the newer ones, and it's up to 6.0 now.
That said, it's more of an anime model, however, it works good as a base if you utilize a refinement workflow with two models. Like, homosimile, and then follow it up with pornmaster, or realitymaster. Realitymaster makes better reality images, though, can sometimes lean female, so.. you may need to experiment around with the model weighting. And pornmaster is better at creating realistic verstions of anime/cartoon characters. In other words, it's more submissive to lora use, and produces more accurate results.
Otherwise, plantmilk, is a really good all around model. Leans anime, but can low-key do realism, though not always, depends on loras used and other settings. Also, it's a good base for use with pornmaster. Can be slightly more accurate, depending on what you're attempting to accomplish.
one of my favorite model for realism, works well with ipadapter too. as usual dpm++2m - karras @25-35 steps works best IMO (intended for newbies that almost always ask for what or how), add in 12-16 steps of 0.3 denoising strength hires fix (use efficiency nodes ksampler's script node)
Z-image is my favorite for now (excluding nudity), it does everything awesome.
"AlbedoBase XL - v3.1-Large is not enabled for generation." š¢
Let's bid it
What kind of vulvas did you train? Almost everything is really great, but they're really not good.
Has anyone been able to reliably generate NSFW with this model??
Its one of the best around for clarity and realism, but even slightly towards NSFW and feels like its all being restricted and have to start adding ":5" etc just to get anywhere...
Yes, it's possible. Steps:+70 will helps you. I generated full nude body+ pubic hair. Clear criticism. No porn. This is one of the best model i have ever seen. Real masterpiece!
I was never able to reliably generate NSFW with this model. However, I kept coming back to this model because the subjects and lighting were so beautiful. Ultimately, I created a checkpoint merge. I merged this model with the NSFW model I liked best (uberRealistic V6) (https://civitai.red/models/790652?modelVersionId=919063) It took a few experiments to find the right ratio, but 60% UberRealistic and 40% AlbedoBaseXL turned out to be the sweet spot for me. That gave me a model that still looked like Albedo but with NSFW capability. Torsos and legs are sometimes elongated, but different seeds will get me the desired result. I generally use Euler_a with Karras, CFG 4
I realize this is an old model by now. If you ever update, do you think you could try to address the long torso issue? A lot of model families from this time (mid-late 2024) had this issue, probably several of the models you merged included this.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.