Important note
I am not claiming this is the best realism model, and that is not its purpose. It works well as a realism model, and will give you the most balanced output of concepts when compared to others, but other, more targeted models, will render specific concepts better than this one will. (eg. models tuned to render only beautiful women will produce them better than this one). This model is designed to have to have a mathematically objective (impartial, unbiased, meaning it doesn't matter how well you or I think it looks) balance of most of the realistic model weights that exits. It was designed primarily to be used as a universal training base for realistic concepts as it will give you the highest chance to create realistic LoRAs that are more consistent across the various realistic models that exist. It will have a more balanced and broad understanding of more concepts than other models without having as many heavy biases get in the way as much, making concept targeted training (such as I do with my sliders) easier to accomplish. Keep this in mind and please don't compare my apples to your favorite oranges. :)
I will try to keep the generation parameters the same on each version. This is for better or worse, since v1 images were heavily curated, sometimes other versions may not produce that exact prompt as well. But I want to make sure the changes in the model are easy to see from each version. I may edit the prompts slightly to prevent newly added nudity (or any other NSFW) or other concepts that may break TOS, as I did in one of the photos in v2. But will do my best to keep them as consistent as possible within those constraints and only modify them as little as necessary. I will leave malformed things malformed. It is what it is.
New in V2
Heavily modified the algorithm to better handle outliers in the weights and prevent them from being amplified.
added weighted penalty to some models that were presenting their features more than they should, still investigating why it happened
increased model count from 42 to 55. I didn't remove any models, but did penalize a few
added 2 versions of an unreleased Bizarre model I have been working on that will help introduce and identify new hand captioned and curiated concepts that do not exist in the current models
About
This is not just another merge. This was done with a custom merging algorithm I have been working. It is a massive, highly computed, merge of 42 55 hand picked models (many of my own, custom trained an unreleased ones) where every single weight, down to even the the layer bias weights, were individually compared to every other model to create a mean squared similarity matrix for each weight to calculate a very precise mixture of the model weights to maintain all of their uniqueness while removing all of the "inbreeding" that results in the same models being merged over and over again.
The result is a model that shares an objective balance of many of the top photorealistic models here while including many unique features and understandings from some of the more obscure ones, as well as many of my private unreleased realistic training runs. I did not plan to release this initially as it was designed to be used as a training base for my LoRA work. But I am frankly in love with it as a go to model, so I wanted to share.
It works great as a training base for realistic LoRAs and because of the balancing algorithm used to calculate the merge, it will give you the best chance of making LoRAs, and other networks and embeddings, that will generally be more compatible with more of the realistic models on here than training on any of them individually.
Weights
Weight breakdowns of averages as well as a link to the complete weight by weight breakdown can be found in the version information for each model version
Thanks
I want to say thank you to all the hard work from all the people who created of all the models that went into this, to everyone at Civitai, and to this community as a whole.
Description
Total models: 55
A complete breakdown of the weight by weight merge weights can be found: https://gist.github.com/jaretburkett/6fba16144939d2b62ea01e2f5803b921
Average weight merge strengths:
FAD-foto-assisted-diffusion_V0: 0.0540084862424792
Ostris_Bizarre_v1: 0.043663373291957026
nextphoto_v30: 0.03743993853748925
movieDiffusionV12_v12safetensors: 0.03600372229449346
amIReal_V4: 0.03378963086737185
Flesh_v0.1: 0.03257149053355277
eroticVision_v4: 0.031695323350022044
Ostris_Bizarre_v2: 0.031132479072094384
darkestModel_v10: 0.029897350966803703
hardblend_: 0.024611756434693727
vodkaByFollowfoxAI_v40: 0.022862042315713992
epicrealism_pureEvolutionV5: 0.022516284124237246
chilloutmix_NiPrunedFp16Fix: 0.022022376616238324
duchaitenRetro_v10: 0.02101808764323454
leosamsMoonfilm_filmGrain20: 0.019389446127881016
acornIsSpinning_photoV2: 0.01938408058846279
lazymixRealAmateur_v10: 0.019136109123272554
buxomBritsV12_v10: 0.018568174938867678
qgoPromptingreal_qgoPromptingrealV1: 0.018015632731100864
photon_v1: 0.016795962746941433
nextphoto_v10: 0.01671644427355322
bondageV11_v11: 0.015833098018201937
cineroSD15_v12BetaFp16: 0.015780317009998976
acornIsSpinning_v30: 0.015243840739637827
cineDiffusion_v3: 0.01522800321490114
juggernaut_v21: 0.014765230566051992
pornmasterAmateur_fp16V5: 0.014573140578796407
analog-diffusion-1.0: 0.01433883833565882
uberRealisticPornMerge_urpmv13: 0.014110604313766067
rmadartSD15_v80: 0.013661339978292547
epicrealism_newAge: 0.013254305495629223
Ostris_Humans_v2.00165000: 0.013007605556069172
projectAIO_v6: 0.012622098326514092
cyberrealistic_classicV16: 0.012474950794721872
absolutereality_v10: 0.012473885372255723
absolutereality_v181: 0.012421313006971484
metagodRealRealism_v10: 0.012368034554514473
metagodSelfieMix_metagodSelfieMix10: 0.01231087465895141
realisticVisionV20_v20: 0.012294294220731165
realisticVisionV51_v51VAE: 0.012223630579707968
Ostris_Real_v2BETA7: 0.012019049160841396
photosomnia_vfinal: 0.0119670890694032
humans_v10: 0.011807964014442727
pureperfection_v20: 0.01175922394335237
icbinpICantBelieveIts_seco: 0.0117064250037581
cyberrealistic_v33: 0.011647563302032055
juggernaut_aftermath: 0.011553722326041885
analogMadness_v50: 0.01138494452968444
lomostyle_v10: 0.011371842939237698
aZovyaPhotoreal_v2: 0.01124316312545857
edgeOfRealism_eorV20Fp16BakedVAE: 0.010866832233069985
Ostris_Real_v1: 0.010751596281342974
lrmLiangyiusRealistic_v15: 0.010726841650227427
realisticVisionV40_v40VAE: 0.01057069511268954
consistentFactorEuclid_euclidV51: 0.010399448307328226FAQ
Comments (13)
In V2 you get much more beer at the same prompt!!
Jokes apart, your work is very appreciated, original and professional, starting from the amazing covers... do you go to sleep sometimes?
Thank you Ostris, best whishes!
I sleep on sundays
@Ostris HAHAHA!!! Ok it's just Thursday... God work to you!
Just one thought... I don't know if it could be reasonable, training speaking... ever thought about an aging slider?
Have a nice day Sir
@vivalabadila I have an aging slider, just currently do not plan to release it. My sliders are not context aware, meaning, whatever that person is doing, they are now doing it with the adjusted concept while prompting will adjust what they are doing. An age slider is desperately needed as prompting for anything other than 18-20 and 65+ is hard. A 35 yr old is hard to get. However, that slider also goes the other way, and i'm not touching that with a 10 foot pole.
This model has some interesting influences, but the subtleties it brings to the table are kind of lost with a pruned/fp16 model. Are you considering a full/fp32 version?
I will check to see if there is much difference. I will have to rebuild it as I save at fp16, though I do all the calculations at fp32. In my experience, there is not enough of a difference to justify the doubled size, but I will check.
@Ostris its mostly placebo that people expecting for
@Ostris The main advantage is that people who are blending with other models or their own to achieve specific results will get degraded results working with a pruned model. I keep my own pet models that I blend checkpoints into, rather than using existing checkpoints as-is. fp16 pruned checkpoints can be massaged back into something useful for such work, but it's certainly not trivial.
@monkeyman007 if you're just prompting-and-going, you're correct. But I've done extensive testing with both raw fp16/32 pruned/full models and with maintaining my own local merges. There is a substantial difference in many use-cases.
It's just placebo that most people don't understand, specially because they generate images using fp16 anyways :)
strange this model dont work for me, the gui i use (nmkd) dont load it at all.
Extremely useful and unique model. I love it's knowledge about different (traditional media) art styles and artists. Pretty surprising feature in a photorealistic model.
This is what SD1.5 base model should've been!
This model doesn't work with the Linaqruf trainer, at least not out of the box. Although it can generate great images, the Linaqruf trainer reports some key mismatches. The stable diffusion toolkit reports corruption in the model.
If Linaqruf or whatever fine Tuner you're using refuses to fine tune this model or it's Lora, then do the following:
In A1111, go to the checkpoint merger. Put SD1. 5 as model A and this one as model B. Set the multiplier to 1. Then do the merge (which should take 0% of SD 1.5 but fix the corrupt layers.) You'll get a new model file which you should select as your base model if you fine tune this or its Lora.
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.