<h3 id="click-on-the-pictures-to-see-them-in-better-quality.">Click on the pictures to see them in better quality.</h3>__________________________________________________________________________________________________90% of The Dataset for Dixar consists of images created by me. About 6k images now.<h3 id="v4-below">V4 Below</h3>A new version already? I know right!Dixar Version 4 is a merge of trained Dixar models, similar to v3. It is less cinematic than v3 but I feel that it is more stable and more faithful to the original Dixar.I was lazy with the example images and for many of them, I just stole prompts from images I liked that you morons made with slight alterations because you all are sick mother fuckers. The Prompts were taken from pictures I liked in both Dixar v2 and v3. I really used minimal effort on the example images and they are only lightly detailed.The Clip is also lightly trained as well, although I just used whatever tags Kohya gave me. I don't really know how to make high quality captions/tags and i'm not tagging 6k images by hand lolFor once the example images actually contain the prompts and shit. For most of the images it was:Euler ABeta40 StepsCFG of 5I used the SeedVR2 node in Comfy to upscale.Idk if it is better, worse, or just different than v3, but the quality looked good enough to publish.<h3 id="if-you-merge-it-with-something-just-credit-me.">If you merge it with something just credit me.</h3>Likely no more updates to Dixar until Z-image base comes out. I will train on that next.Post your images and maybe i'll use your prompt for Dixar five's showcase images.That's kinda it, by now you should all know how to use an image generation model.PS- The main picture that is always in front (punk silver haired girl with backwards hat) got flagged for review for some reason so after it got approved by the mods the day after I posted this, i moved it to image 5.Have a wonderful day and go fuck yourself.__________________________________________________________________________________________V3 Below<ul><li>Dixar version 3 is a trained model...mostly. It is a merge of five separate trained versions of Dixar.</li></ul>Each version was trained on Kohya. I am still learning how to train checkpoints and honestly, I kinda suck at it. I have trained it about 20 separate times and I have six versions that I like but none are perfect as standalone models. This is probably among the best I have so far. The dataset at the time consisted of 2600 images, 90% of the dataset was created by me, which is where I think the model is unique, very high quality images in a dataset, almost all of which were created by me. They were created using everything you can imagine, some in Comfy with various versions of Dixar that I have, some using Chat GPT, Some using Nano Banana Pro, Hunyuan image 3, Wan2.5 image, Seed Dream 4, Qwen, and Midjourney.The 10% of the images in the dataset that were not created by me are screenshots from 3D movies. However, they are not the screenshots you would find online. All of the screenshots in the dataset were taken by me because I didn’t like the quality you can find online and I have a true 4K monitor. Every screenshot was also hand edited by me. And yes, I had to watch every 3D movie to take the screenshots.Just building the dataset probably cost a few thousand $$$ worth of gpu time or credits on various sites, as well as a ton of time. Currently I have brought the dataset up to 6910 images but I have not trained, or even organized those images yet.This version defiantly has some issues, but it is good enough to post.It aims to be closer to a screenshot from a Pixar movie, it's obviously not there yet but it is getting there.You don't have to specify 3d, Pixar, or anything but I feel it looks better when you do, just like anything with image generation you are telling the model what you want it to make for you.Euler ABeta or SGM50 Steps3 CFGIt does not know specific characters, although I only tried one.It is also fragile and easily breakable. It will work with some loras but others just break it.Some of the example images are hidden if you have X and XXX enabled.The very last image is a raw batch of 50 images.<h2 id="just-credit-me-if-you-merge-it.">Just credit me if you merge it.</h2>I Like to see the images you create, plus if I like them a lot, I'll add them to my master dataset for v4.Next, I will train my anime model, Hemorrhoids, but I must build my own dataset first though.<h3 id="as-always-have-a-wonderful-day-and-go-fuck-yourself.">As always, have a wonderful day and go fuck yourself.</h3>__________________________________________________________________________________________________------------------------------------V2 Below-----------------------------------------------------Dixar 2: Double the DixLow CFG suggested 3 to 3.5Sample pics were all made with Deis and SGMI merged Dixar with CyberIllustrious40 + MMMix + A cool 3d checkpoint I found on a Chinese website + an unreleased blender style model I made (not releasing that one yet, it is still not up to my standards).It was also merged with a Lora I made that took 100 of the most 3d pictures I used with 3dish lora for flux. There was no trigger word in there, the images weren't even tagged or captioned. All the pictures in that Dataset were made by me. That lora took 2 months for thousands of images with flux and thousands of individual fixes.... all that for 150 downloads :(Because The above source images in that Lora were made with Flux, it is pretty blush heavy. I do recommend using (Blush:1.2) in the negative prompt.The clip is a 50/50 mix of Dixar and Cyber.<h2 id="there-were-more-sample-images-but-they-got-flagged-for-review-and-i-am-too-lazy-to-make-more.">There were more sample images but they got flagged for review and I am too lazy to make more.</h2>Always remember, I do not care about your opinion.Keep in mind that I have no idea what I am doing. I make models the same way I passed algebra in high school, the guess and check method. Fuck you Mx+B=DickThe very first image was made with the same prompt as image 2 from v1. That prompt is at the bottom. If you know how to read it will be helpful to look at it.For it's first eligible week 6/5/25 to 6/12/25, I'll put the rest of my buzz on it to enable it for sitewide generation. I have 10k left which might be enough to crack the top 200I appreciate everyone who donates BUZZ to me but honestly, I don't need the money. Save the buzz, give it to somebody else, or just bid on this for onsite generation so everybody can use it. Seriously, thank you, but I really don't need the money, I just do this shit as a hobby.It is late and I can't think of anything witty so just go fuck yourself.P.S- Dixar is bad at training loras. Don't waste your bus to train on it______________________________________________________________***V1 Below***I like the name.I don't actually remember everything I merged but it was quite a bit. The model produces the type of images you see.There is a vae baked but feel free to use your own.I did not intend on this model producing NSFW content, but, it can, pretty well too, even when you don't want it to. Just be clear with your prompts. Put what you want the character to wear or it might just give you an image with no clothing.Might need to prompt the skin color sometimes.Use teeth in the negative prompt if it keeps giving you open mouth pictures.When I made this, I did not use anything Disney or Pixar related, it just happens to have a resemblance. None of my prompts had the word Disney or Pixar in it ether, just "3d." I honestly didn't even try it with the word Disney or Pixar in my prompts. I am being serious.Use any sampler/scheduler you want. I am use to flux and my favorite sampler/scheduler with that was Deis and SGM so that is what I tested it on.-No other loras were used in the sample images but they were run through detailers, they were also run through your mom.-The shiny plastic like skin, mostly came from the detailers, not the initial generations.-if some of the images look over sharpened, that is my fault not the models, I am a few IQ points above mentally challenged.If you have any questions or comments, just remember, your opinion is meaningless to me and I do not care.<h3 id="i-found-one-of-my-example-prompts-this-was-image2:">I found one of my example prompts, this was image#2:</h3>Very high studio quality 3d anime girl, 1girl, dressed like a punk skater girl, big smile, backwards hat, very long silver hair, best quality, score_9_up, score_8_up, cinematic city background, spectacular shadows, vibrance, very aesthetic, high resolution, absurdres, depth of field, thick plastic skin, perfect anatomy, very awa.<h3 id="as-always-have-a-wonderful-day-and-go-fuck-yourself.">As always, have a wonderful day and go fuck yourself.</h3>

Dixar

Dixar version 3 is a TRAINED MODEL

dixar_IIITripleTheD.safetensors

Dixar III: Triple the D

Dixar 4: Dix Galore

Dixar 2: Double the Dix

v1.0

Dixar 2: Double the DixLow CFG suggested 3 to 3.5Sample pics were all made with Deis and SGMI merged Dixar with CyberIllustrious40 + MMMix + A cool 3d checkpoint I found on a Chinese website + an unreleased blender style model I made (not releasing that one yet, it is still not up to my standards).It was also merged with a Lora I made that took 100 of the most 3d pictures I used for with 3dish lora for flux. All the pictures in that Dataset were made by me. That lora took 2 months for thousands of images with flux and thousands of individual fixes.... all that for 150 downloads :(The clip is a 50/50 mix of Dixar and Cyber.There were more sample images but they got flagged for review and I am too lazy to make more.Always remember, I do not care about your opinion.It is late and I can't think of anything witty so just go **** yourself.

Dixar2: Double the Dix

Dixar version 3 is a trained model...mostly. It is a merge of five separate trained versions of Dixar.

Each version was trained on Kohya. I am still learning how to train checkpoints and honestly, I kinda suck at it. I have trained it about 20 separate times and I have six versions that I like but none are perfect as standalone models. This is probably among the best I have so far. The dataset at the time consisted of 2600 images, 90% of the dataset was created by me, which is where I think the model is unique, very high quality images in a dataset, almost all of which were created by me. They were created using everything you can imagine, some in Comfy with various versions of Dixar that I have, some using Chat GPT, Some using Nano Banana Pro, Hunyuan image 3, Wan2.5 image, Seed Dream 4, Qwen, and Midjourney.

Just building the dataset probably cost a few thousand $$$ worth of gpu time or credits on various sites, as well as a ton of time. Currently I have brought the dataset up to 6200 images but I have not trained, or even organized those images yet.

This version defiantly has some issues, but it is good enough to post.

It aims to be closer to a screenshot from a Pixar movie, it's obviously not there yet but it is getting there.

You don't have to specify 3d, Pixar, or anything but I feel it looks better when you do, just like anything with image generation you are telling the model what you want it to make for you.

Euler A

Beta or SGM

50 Steps

3 CFG

It does not know specific characters, although I only tried one.

It is also fragile and easily breakable. It will work with some loras but others just break it.

If you are seeing this within a few days of me posting the checkpoint, like half the example images got flagged for review.

Just credit me if you merge it

As always, have a wonderful day and go fuck yourself.

Files with SHA256: 9ba9aa0034ad6690220e70238ae58dd8a6667dd6654bb6f39e1f9dcf9ac603ca