<h2 id="heading-834">V4 is out!</h2>This version uses the same models as V3 (with updated epoch versions) while adding a new base, <a rel="ugc" href="https://civarchive.com/models/4823?modelVersionId=15236">Deliberate V2</a>! Deliberate and 526Mix are combined with 67% 526Mix and 33% Deliberate average weight merge, and then used just like 526Mix was used in previous versions. Additionally, this mix used block weight merging in Comfy to merge together the e6-laion + Polyfur part with the FluffyRock NoPE part at 0.5 CLIP merge, and the result is train differenced onto P.A.W.F.E.C.T just like V3. One last thing, the name is now officially changed to FluffyMix to make it easier to remember.V3 is out!This version uses fluffyrock-NoPE-e144-terminal-snr-vpred-e63.safetensors, e6-laion-576-704-832-lion-e81-terminal-snr-vpred-e30.safetensors, polyfur-lion-e87-terminal-snr-vpred-e36.safetensors, pawfect-alpha-e48-None.safetensors, and v1.5 of 526Mix. e6-laion operates similarly to Polyfur and includes the same MiniGPT4 captioned photographs, but e6-laion is also trained on anime style images from booru sites. e6-laion and Polyfur are merged together at a 0.5 weight average merge for the purposes of this mix. FluffyRock NoPE also operates similarly to FluffyRock Vpred, and while both models use the same dataset and are both v-prediction, FluffyRock NoPE messes with the text interpretation part of its model in order to remove positional encoding, meaning it doesn't require a token limit. It isn't known if this is passed on through merging, however NoPE did have a different and more preferable aesthetic from its sibling, which is why it was chosen. Finally, P.A.W.F.E.C.T is another v-prediction model trained on images from FurAffinity. The unique aspects of this model are its aesthetic and tags, as FurAffinity tags are often diverse and sparse, meaning images are often incompletely tagged and sometimes use uncommon tags. This means P.A.W.F.E.C.T both provides unique tagging and a unique aesthetic to this mix.V2 is out!This version uses polyfur-lion-e76-terminal-snr-vpred-e25.safetensors and fluffyrock-576-704-832-960-1088-lion-low-lr-e126-terminal-snr-vpred-e99 mixed with 526Mix to provide the model with even better knowledge and aesthetics. Polyfur is special, as it attempts to include both high quality photographs autocaptioned by MiniGPT4 in its dataset alongside furry images, so that you get the best of both worlds. The model is early in its lifespan, but contains enough specialised knowledge that it significantly improves the quality of this mix.Unofficial merge of fluffyrock-576-704-832-960-1088-lion-low-lr-e126-terminal-snr-vpred-e99 and <a target="_blank" rel="ugc" href="https://civarchive.com/models/15022/526mix-v14">https://civarchive.com/models/15022/526mix-v14</a> using train difference with the supermerger extension in automatic1111. Train difference is explained here <a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-supermerger/blob/main/calcmode_en.md#train">https://github.com/hako-mikan/sd-webui-supermerger/blob/main/calcmode_en.md#train</a>, and is what allows this model to exist. TLDR, this model uses the a1111 supermerger extension to merge 526Mix with FluffyRock vpred using these settings: 526Mix is set as model A, fluffyrock vpred as model B, v1-5-pruned (sd base model) as model C, add difference as the merge mode, train difference as the calculation mode, alpha is set to 1, and the output is saved as fp16 safetensors.This model requires a config and CFG Rescale. Download the config file and put it next to the regular model in your stable diffusion models folder. If you use a1111, install the CFG Rescale extension here: <a target="_blank" rel="ugc" href="https://github.com/Seshelle/CFG_Rescale_webui">https://github.com/Seshelle/CFG_Rescale_webui</a>. If you use comfy, download the sampler_rescalecfg file from here <a target="_blank" rel="ugc" href="https://github.com/comfyanonymous/ComfyUI_experiments">https://github.com/comfyanonymous/ComfyUI_experiments</a> and put it in your custom nodes folder. Recommended settings are 7.5 for CFG and 0.7 for CFG Rescale.Credit goes to zatochu on the furry diffusion discord for figuring this out, and credit to 526christian for creating the 526mix and lodestone for creating fluffyrock.

FluffyMix

This version uses e76 of Polyfur, e126 of FluffyRock, and update 1.5 of 526Mix.

fluffymix_v20.yaml

fluffymix_v20.safetensors

v2.0

v4.0

v3.0

v1.0

This version uses e126 of FluffyRock Vpred and update 1.5 of 526Mix.

fluffymix_v10.safetensors

fluffymix_v10.yaml

<h2 id="heading-8">This model works correctly only with <a target="_blank" rel="ugc" href="https://civarchive.com/api/download/models/189669?type=Config">config file</a>. Place this file in the same folder as this model.</h2>Fluffy Universe (old name Furry DreamShape) is my personal staple mix that I update to suit my needs, and I do LORA training on it as well.This model merged with such well-known models as YiffyMix, DreamShape, BB95, Koji, epic realism and the core of this model is the best model today in my opinion - FluffyRock As for samplers, I most often use the most powerful "Restart". I experiment a little less often with various subtypes of DPM++ 3M SDE. <a target="_blank" rel="ugc" href="https://civarchive.com/models/92527/fluffyrock-pre-3m-autocomplete">Tag Autocomplete File</a> - courtesy of FluffyRock's tag database. This model is universal and suitable for generating a variety of objects, including realism.<h2 id="heading-6"> I express special gratitude to each author of the listed models present in my mix.</h2><h2 id="heading-7"> LORA training on models with the FluffyRock base includes the following instructions:</h2><ol><li>Must be activated "parameterization v". This setting is located in the tab when selecting the base model. The program may display an error that this model does not support this function. Ignore the warning</li><li>it is advisable to use the parameter "noise_offset" around 0.1-0.9.</li></ol><h2 id="heading-13"> Reminder</h2>Model 1.0 is my old mix without FluffyRock. However, this mix is ​​useful when working with common lore.LoRAs trained on non-vpred FR models may perform significantly worse on newer versions my mix.

Fluffy Universe (Furry)

Furry DreamShape is my personal staple mix that I update to suit my needs, and I do LORA training on it as well. When installing this Furry DreamShape, do not forget to install the configuration files that must be placed in the folder with this model.After figuring out the settings, I was able to combine the EasyFluff model with my existing merge. The results were incredible. Many flaws such as "burning colors" were corrected, and the overall level of detail increased significantly. Since EasyFluff is by far the model with the largest dataset, I will be building my model on its basis in the future. As for the rest of the description, this model also includes BB95, some anime models, YiffMix, and Dreamshape. In the future, if there will be updates to the main custom models, like the previously mentioned Dreamshape, my model will also be updated. The same goes for the main EasyFluff model. As for the recommended samplers, I almost always use the latest DPM++ 3M SDE, experimenting with their subspecies. All the tags specified in the YiffMix and FluffyRock datasets work on this model as well.

fluffyUniverseFurry_v20Vrped.safetensors

fluffyUniverseFurry_v20Vrped.yaml

v2.0 Vrped

Fluffy Universe 1.0 Vpred

v2.1 Vpred

v2.0 (Pruned) Vpred

v1.0 (Non Vrped)

fluffyUniverseFurry_fluffyUniverse10.safetensors

fluffyUniverseFurry_fluffyUniverse10.yaml

<h3 id="heading-86">March 29th 2024 Update: A more experimental personal merge based off of Fluffusion R3 has been posted. This merge is different from previous versions of this model series. As such it will not be explicitly given that naming scheme. As it's v-prediction, remember the yaml or sampling nodes. Fluffusion is trained with clip skip 2, so make sure you set that in your UI of choice.</h3><h1 id="heading-87">Below is information that is outdated. That is all.</h1><h2 id="heading-9">Results oversaturated, smooth, lacking detail? No result at all? Please read the following!</h2><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0c3836b0-34be-4362-8115-7eb37000658b/width=525/0c3836b0-34be-4362-8115-7eb37000658b.jpeg" /><h2 id="heading-6">This is a terminal-snr-v-prediction model and requires a configuration file placed alongside the model and the CFG Rescale webui extension! See further down for more information and equivalent for ComfyUI.</h2><h1 id="heading-1987">HOW TO RUN THIS MODEL</h1>- This is a terminal-snr-v-prediction model and you will need an accompanying configuration file to load the model in v-prediction mode. You can download it in the files drop-down and place it in the same folder as the model.- You will also need <a target="_blank" rel="ugc" href="https://github.com/Seshelle/CFG_Rescale_webui">https://github.com/Seshelle/CFG_Rescale_webui</a>. This extension can be installed from the Extensions tab by copying this repository link into the Install from URL section. A CFG Rescale value of 0.7 is recommended by the creator of the extension themself. The CFG Rescale slider will be below your generation parameters and above the scripts section when installed. If you do not do this and run inference without CFG Rescale, these will be the types of results you can expect per this <a target="_blank" rel="ugc" href="https://arxiv.org/pdf/2305.08891.pdf">this research paper</a>.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/aaef6b3f-8cde-4a34-a4ae-6b7a066a3766/width=525/aaef6b3f-8cde-4a34-a4ae-6b7a066a3766.jpeg" />- If you are on ComfyUI, you will need the sampler_rescalecfg.py node from <a target="_blank" rel="ugc" href="https://github.com/comfyanonymous/ComfyUI_experiments">https://github.com/comfyanonymous/ComfyUI_experiments</a>.<h1 id="heading-1988">HOW TO PROMPT THIS MODEL</h1>- Natural language has a strong impact on this model as of V7. It won't understand everything but it can get you places. And then when you can't squeeze any more out of it, you can use e621 tags to help refine your prompt. (or you can just use tags, either way will work.)- Start your prompt with a short natural language prompt describing what you want, then pad it with e621 tags to refine specific concepts. As this is a v-prediction model, prompt interpretation can be a bit more literal.- Many flavor words and artists from SD 1.5 work again.- PolyFur is trained on MiniGPT-4 captions, so try being really flowery with your prompts and use sentences even. (check the generation info on the preview images to see how)- If an established character isn't coming out accurately, try increasing the strength of their token and adding a few implied tags that describe their appearance. Keep in mind that characters that aren't very popular or don't have many images in FluffyRock's dataset typically won't fare that well without a LORA.- Avoid weighting camera angle keywords too strongly, especially <code>close-up</code>.- Resolutions between 576 and 1088 should work reasonably well as that is the range of FluffyRock.<h1 id="heading-1989">Model Info and Creation Process</h1>V8.1 now available. Last major version for a while. Currently the quality tag LORA from Feffy is not baked in and I recommend using it as a LORA. As of this V4-V6, @Feffy's Quality Tag LORA was baked into the FluffyRock part of this model. Add masterpiece, best quality and worst quality, low quality, normal quality to your positive/negative prompt.This model is a "Train Difference" mix (see <a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-supermerger/blob/main/calcmode_en.md#traindifference">https://github.com/hako-mikan/sd-webui-supermerger/blob/main/calcmode_en.md#traindifference</a>) of FluffyRock and PolyFur onto a diverse and versatile base model to expand the text encoder's range of prompting and understanding of concepts outside of e621 tags. This process differs from traditional merging and block merging and shouldn't be confused with "Add Difference" that we've had. The current base model is a mix of 6 non-furry models at varying ratios, which are mostly merges as well that cover a lot of ground in terms of prompting capability. As of V6 the process has gotten more intricate with CLIP tweaks and block merging.<h3 id="heading-7">Credits go to Lodestone Rock and company for training such an expansive model. <a target="_blank" rel="ugc" href="https://civarchive.com/models/92450?modelVersionId=124661">https://civarchive.com/models/92450?modelVersionId=124661</a> <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/furryrock-model-safetensors/tree/main">https://huggingface.co/lodestones/furryrock-model-safetensors/tree/main</a></h3><h3 id="heading-8">And pokemonlover69 over at <a target="_blank" rel="ugc" href="https://civarchive.com/models/124655?modelVersionId=136127">https://civarchive.com/models/124655?modelVersionId=136127</a> for humoring my insanity on this process.</h3><h3 id="heading-9">And Feffy for their Fluffyrock Quality Tags LORA. <a target="_blank" rel="ugc" href="https://civarchive.com/models/127533">https://civarchive.com/models/127533</a></h3>

EasyFluff

<ul><li>Fixes oversaturated and burnt results when using shorter prompts.</li><li>Should provide generally cleaner results and less mangled low-level concepts.</li><li>Artist tags should behave better without overpowering the output with little to no emphasis.</li></ul>

easyfluff_v81.yaml

easyfluff_v81.safetensors

V8.1

5000-defaultVAE(FFR3-CS2)

V10-PreRelease

More up to date and experimental versions available at: <a target="_blank" rel="ugc" href="https://huggingface.co/zatochu/EasyFluff">https://huggingface.co/zatochu/EasyFluff</a>Biggest change is more capability of photorealism when prompted. Outputs should be reasonably similar to V9's. Natural language may be slightly weaker and more nuanced prompts may not work as well as before. Results with Euler A as the sampler may be susceptible to worse eyes and finer details but not thoroughly tested. If you're finding results over-saturated or too vibrant, consider using the VAE available here. <a target="_blank" rel="ugc" href="https://civitai.com/models/118561?modelVersionId=131656">https://civitai.com/models/118561?modelVersionId=131656</a>

easyfluff_v10Prerelease.safetensors

easyfluff_v10Prerelease.yaml

This checkpoint was trained on 3.3m images of normal to hyper sized anime characters. It focus mainly on breasts/ass/belly/thighs, but now handles more general tag topics as well. It's about 50%/50% anime and furry images as of v8. See change log article below for more version details, and future plans. Note: This will be my final SD1x model. I wanted to see what the hyperfusion dataset was really capable of on sd1.5. So I let it train on 2x3090s for 10 months to squeeze every bit of concept knowledge out of it. This is the best concept model Ive trained so far, but it still has the usual SD1x jankiness. I probably kept the Text Encoder LR too high for too long (0.5x -&gt; 0.3x).Big shoutout to <a target="_blank" rel="ugc" href="http://stuffer.ai">stuffer.ai</a> for letting me host my model on their site to gather feedback. It was critical for resolving issues with the model early on, and a great way to see what needed improvement over time. V9 is a v_pred model, so you will need to use the YAML file in A1111, or the vpred node in Comfy along with cfg_rescale=0.6-0.8 in both. A1111 will also need the <a target="_blank" rel="ugc" href="https://github.com/Seshelle/CFG_Rescale_webui">CFG_Rescale extension</a> installed.I posted one old example using ComfyUI workflow here <a target="_blank" rel="ugc" href="https://civarchive.com/images/64978187">https://civarchive.com/images/64978187</a> Other links:The OG hyperfusion LoRAs can be found here <a target="_blank" rel="ugc" href="https://civarchive.com/models/16928">https://civarchive.com/models/16928</a> Also a back up <a target="_blank" rel="ugc" href="https://huggingface.co/thojm/HyperFusion">HuggingFace link</a> for these modelsUploaded 1.4 million custom tags used in hyperfusion <a target="_blank" rel="ugc" href="https://huggingface.co/datasets/thojm/hyperfusion_classified_tags_export">here </a>for integrating into your own datasets<h2 id="changelog-article-link"><a target="_blank" rel="ugc" href="https://civarchive.com/articles/3320">Changelog Article Link</a></h2><h3 id="recommendations-for-v9_vpred-finetune:">Recommendations for v9_vpred finetune:</h3>sampler: Anything that is not a Karras sampler. Don't use Karras! Training with --zero_terninal_snr makes that sampler problematic. Also you will need to use the uniform scheduler in A1111, or "simple,normal" in Comfy at leastnegative: I tested each of these tags separately to make sure they had a positive effect:worst quality, low rating, signature, artist name, artist logo, logo, unfinished, jpeg artifacts, artwork \(traditional\), sketch, horror, mutant, flat color, simple shadingpositive: "best quality, high rating" for the base style I trained into this model, more details in Training Data docscfg: 7-9cfg_rescale: 0.6-0.8 rescale_cfg is required for this v_pred model. lower values tend to have less body horror, but darker images.resolution: 768-1024 (closer to 896 for less body horror)clip skip: 2zero_terminal_snr: Enabledstyling: You will want to choose a style first. The default style is pretty meh. Try the new artist tags included in v8+, all tags can be found in the tags.csv by searching for "(artist)". See example images for art styles.Lora/TI: loras trained on other models will not work with this model, even loras trained on other v_pred models are not guaranteed to work here.<h3 id="recommendations-for-v8-lora:">Recommendations for v8 LoRA:</h3>sampler: Anything that is not a Karras sampler. Don't use Karras! Training with --zero_terninal_snr makes that sampler problematic.Lora/TI: If you are using LoRA's/TI's trained on NovelAI based models, they might do more harm than good. Try without them first.negative: low rating, lowres, text, signature, watermark, username, blurry, transparent background, ugly, sketch, unfinished, artwork \(traditional\), multiple views, flat color, simple shading, unfinished, rough sketchcfg: 8 (it needs less than LoRA hyperfusion) resolution: 768-1024 (closer to 768 for less body horror)clip skip: 2styling: Try the new artist tags included in v8, all tags can be found in the tags.csv by searching for "(artist)"<h3 id="tag-info-(you-definitely-want-to-read-the-tag-docs-see-:training-data)"> Tag Info (You definitely want to read the tag docs, see :Training Data)</h3> Because hyperfusion is a conglomeration of multiple tagging schemes, I've included a tag guide in the training data download section. It will describe the way the tags work (similar to Danbooru tags), which tags the model knows best, and all my custom labeled tags. For the most part you can use a majority of tags from Danbooru, Gelbooru, r-34, e621, related to breasts/ass/belly/thighs/nipples/vore/body_shape.The best method I have found for tag exploration is going to one of the booru sites above and copying the tags from any image you like, and use them as a base. Because there are just too many tags trained into this model to test them all.<h3 id="tips">Tips</h3><ul><li>Because of the size and variety of this dataset, tags tend to behave differently than most NovelAI based models. Keep in mind your prompts from other models, might need to be tweaked.</li><li>If you are not getting the results you expect from a tag, find other similar tags and include those as well. I've found that this model tends to spread its knowledge of a tag around to other related tags. So including more will increase your chances of getting what you want.</li><li>Using the negative "3d" does a good job of making the image more anime like if it starts veering too much into a rendered model look.</li><li>Ass related tags have a strong preference for back shots, try a low strength ControlNet pose to correct this, or try one or more of these in the negatives "ass focus, from behind, looking back". The new "ass visible from front" tag can help too.</li><li>...more tips in tag docs</li></ul><h3 id="extra">Extra</h3> This model took me months of failures and plenty of lessons learned (hence v7)! I would eventually like to train a few more image classifiers to improve certain tags, but all future dreams for now.As usual, I have no intention of monetizing any of my models. Enjoy the thickness!<hr /><h3></h3>-Tagging-The key to tagging a large dataset is to automate it all. I started with the wd-tagger (or similar danbooru tagger) to append some common tags on top of the original tags. Eventually I added an e621 tagger too, but I generally only tag with a limited set of tags and not the entire tag list (some tags are not accurate enough). Then I trained a handful of image classifiers like breast size, breasts shape, innie/outie navel, directionality, motion lines, and about 20 others..., and let those tag for me. They not only improve on existing tags, but add completely new concepts to the dataset. <s>Finally I converted similar tags into one single tag as described in the tag docs</s> (I stopped doing this now. With 3m images it really doesn't matter as much).Basically any time I find its hard to prompt for a specific thing, I throw together a new classifier, and so far the only ones that don't work well are ones that try to classify small details in the image, like signatures.Starting in v9 I will be including ~10% captions along side the tags. These captions are generated with CogVLM.I used this to train my image classifiers <a target="_blank" rel="ugc" href="https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification">https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification</a>Ideally, I should train a multi-class-per-image classifier like the Danbooru tagger, but for now these single class-per-image classifiers work well enough.-Software/Hardware-The training was all done on a 3090 on Ubuntu. The software used is Kohya's trainer, since it currently has the most options to choose from.

hyperfusion_vpred finetune 3.3m images

This version of hyperfusion was trained on 3.3 million images over 10 months, and is a v_prediction + zero_snr model based on SD1.5. This version was trained on SD 1.5, so there is no NovelAI influence in this checkpoint.More image classifiers trained, and existing classifiers improved (list of classified tags under Training Data section)Training Notes:<ul><li>~3.3m images</li><li>LR 4e-6</li><li>TE_LR 1e-6, droped to 1e-7 (after epoch 10)</li><li>batch 8</li><li>GA 16</li><li>2x3090s so 2x the base batch size. total v_batch = 256</li><li>total images seen: 190_000 * 256 = 48_000_000</li><li>AdamW-8bit (ADOPT for the last epoch as a test)</li><li>scheduler: linear</li><li>base model SD1.5</li><li>No custom VAE, usually use the original SD1.5 VAE</li><li>flip aug</li><li>clip skip 2</li><li>525 token length (appending captions + tags made this necessary)</li><li>bucketing at 768 max 1024<ul><li>bucket resolution steps 32 for more buckets</li><li>trained at 768 for the first 10 epochs, and 1024 for the last 6</li></ul></li><li>tag drop chance 0.15</li><li>caption_dropout 0.1</li><li>tag shuffling</li><li>--min_snr_gamma 3</li><li>--ip_noise_gamma 0.02</li><li>--zero_terninal_snr</li><li>about 10 months training time</li></ul>Custom training configs:I have implemented a number of things into Kohys's training code that have been suggested to improve training, and kept the things that seemed to make improvements.<ul><li>drop out 75% of tags 5% of the time to hopefully improve short tag length results</li><li>soft_min_snr instead of min_snr</li><li>--no_flip_when_cap_matches: Prevent flipping images when certain tags exists like "sequence, asymmetrical, before and after, text on*, written, speech bubble" etc... This should help with text, and characters with asymmetrical features.</li><li>--important_tags: move important tags to the beginning of the list, and sort them separately from the unimportant ones (suggested from NovelAI if I remember correctly).</li><li>--tag_implication_dropout: Dropout similar tags to prevent the model from requiring them both to be present when generating. Like "breasts, big breasts" breasts will be dropped out 30-50% of the time. I used the tag implications csv from e621 as a base and added tags as needed. Even with 10%-15% tag dropout, some tag pairs were still being associated too often, this definitely made a difference. I think there were about 5k tags in total on the dropout list.</li><li>12% of the dataset is captioned with CogVLM, as well as cleaning up many of the captions with custom scripts that correct common problems.</li><li>Tags vs Captions: 70% of the time use tags, ~20% of the time use captions (if they exist), 10% of the time combine tags with captions in different orders.</li></ul>If I remember more custom changes, ill add them later.

hyperfusionVpred_v9Vpred_trainingData.zip

hyperfusionVpred_v9Vpred.safetensors

hyperfusionVpred_v9Vpred.yaml

v9 vpred

a high quality mixV2 needs CFG rescale uimodels used:yiffymix 34yiffymix 31fluffusion r1unico-something-furryrealistic vision v5.1i recommend UniPc sampler, 30 steps and 7.5 cfg

ARYM - A Random Yiff Mix

old mix i had but works better than 1.0

arymARandomYiffMix_v20.yaml

arymARandomYiffMix_v20.safetensors

fluffyUniverseFurry_v21Vpred.safetensors

fluffyUniverseFurry_v21Vpred.yaml

This version uses v1.5 of 526Mix, v2 of deliberate, e156 of FluffyRock NoPE, e86 of e6-laion, e93 of Polyfur, and e50 of P.A.W.F.E.C.T

fluffymix_v40.safetensors

fluffymix_v40.yaml

easyfluff_v6.safetensors

easyfluff_v6.yaml

Uses update 1.5 of 526Mix, e144 of FluffyRock NoPE, e81 of e6-laion, e87 of Polyfur, and e48 of P.A.W.F.E.C.T.

fluffymix_v30.safetensors

fluffymix_v30.yaml

<ul><li>fix washed out colors</li><li>fix lost and washed out backgrounds</li><li>fix dilution of concepts that occured in V8.1</li></ul>

easyfluff_v9.safetensors

easyfluff_v9.yaml

fluffyUniverseFurry_v20PrunedVpred.safetensors

fluffyUniverseFurry_v20PrunedVpred.yaml

V7 should be a significant improvement from V6. Concepts should be more accurate and natural language prompts should work even better.

easyfluff_v7.safetensors

easyfluff_v7.yaml

One of my friends liked it so I'm uploading it for him. It's a "trained difference" merge of Fluffyrock low lr e233 terminal snr vpred e206 with helloYoung25d v1.0 and another model. The intention was to get a sort of Disney style but I think it's pretty versatile with the artist tags. Less furry focused than Fluffyrock but should still be able to generate anthros.USE E621 TAGSCopied from the EasyFluff page:- This is a terminal-snr-v-prediction model and you will need an accompanying configuration file to load the model in v-prediction mode. You can download it in the files drop-down and place it in the same folder as the model.- You will also need <a target="_blank" rel="ugc" href="https://github.com/Seshelle/CFG_Rescale_webui">https://github.com/Seshelle/CFG_Rescale_webui</a>. This extension can be installed from the Extensions tab by copying this repository link into the Install from URL section. A CFG Rescale value of 0.7 is recommended by the creator of the extension themself. The CFG Rescale slider will be below your generation parameters and above the scripts section when installed.Highly recommend <a target="_blank" rel="ugc" href="https://civarchive.com/models/127533?modelVersionId=262966">Fluffyrock Quality Tags LORA</a> and the <a target="_blank" rel="ugc" href="https://civarchive.com/models/87781?modelVersionId=94126">boring_e621 embedding</a>Recommended settings:<ul><li>Resolution: 832x832 (really tentative here, anything from 576 to 1088 should work)</li><li>Sampling method: Euler a</li><li>Schedule type: Automatic</li><li>Sampling steps: 44</li><li>CFG Scale: 7</li><li>CFG Rescale: 0.7</li><li>Quality keywords: masterpiece, (best quality, high quality:1.2)</li><li>Negatives: boring_e621_v4, (worst quality, low quality, normal quality:1.2), bad-hands-5, lowres, bad anatomy, bad hands, text, error, missing fingers, blurry,</li></ul>

Fluffyrock Unfluffed

fluffyrockUnfluffed_v10.safetensors

fluffyrockUnfluffed_v10.yaml

ADHDMix is an EasyFluff-style model merge based on the <a rel="ugc" href="https://huggingface.co/lodestones/furryrock-model-safetensors/tree/main">fluffyrock</a> model family.For the nerds:The UNET is created by combining the following models:Train Difference with 30% alphaModel A: Fluffyrock Unleashed V1 (FRU)Model B: Fluffyrock Polyfur e142 (PF)Model C: Fluffyrock v-pred e257This model (meth) is then added via weight sum onto <a target="_blank" rel="ugc" href="https://civarchive.com/models/4384/dreamshaper">Dreamshaper 8</a> (DS). For the CLIP, a merge a weight sum of 85/15 FRU/PF is weight summed with 10% DS, and then this is added onto the existing clip (meth) with a ratio of 45/65 intermediate/existing.

Files with SHA256: cb020f82ffb3c62f70b0ed9fbc8e9bf18d95cd5588375d815799e717d915b2a4