File: 1777245489549014.jpg (2.2 MB)
aia/ - AI Art Thread #78
180 days until Halloween Edition.
▶WARNING
>By posting in these threads, you recognize that AI generation is a valid artistic process.
Please respect AI artists like any other artist.
Backends:
>https://github.com/LykosAI/StabilityMatrix
This is more of a one-click manager that you can install / run the other backends though. Easiest by far. Recommended to download Forge through it and use its built-in Civitai browser to download models with.
For models not through the in-built downloader just download and drop them in Stability_Matrix\Data\Models\* (lora folder if a lora, TextualInversion if embedding, stablediffusion if checkpoint...)
>reForge
https://github.com/Panchovix/stable-diffusion-webui-reForge
I would recommend installing this through Stability Matrix.
>ComfyUI
https://github.com/comfyanonymous/ComfyUI?tab=readme-ov-file#installin g
Has a learning curve but is extremely customizable and usually has the latest methods / papers implemented first.
>"But I don't have a decent GPU"
NovelAI also does great pony: https://novelai.net
You could also use a service like civitai.
Models:
>Pony Diffusion V6 XL
https://civitai.com/models/257749/pony-diffusion-v6-xl?modelVersionId= 290740
>NoobAI-XL Great new alternative. Is closer to novelai than anything else. Really diverse yet coherent positioning / styles, knows artist tags.
https://civitai.com/models/833294/noobai-xl-nai-xl
>ZoinksNoob, a popular new model built off ChromaXL Mix, which itself was built off of NoobAI. Very, very close to NovelAI and arguably better at many things, it does ponies very well and has much less of a "burnt-in" default style than something like Pony Diffusion V6 does.
https://huggingface.co/zatochu/ZoinksNoob
>Purplesmart.ai best for starters and with a great community
https://purplesmart.ai
>Useful artist style LoRA:
https://civitai.com/models/317578/pdv6xl-artist-tags?modelVersionId=35 6175
>LoRAs by /mlp/:
https://rentry.org/ponyxl_loras
>LoRAs by /h and other useful info:
https://rentry.org/ponyxl_loras_n_stuff
>How to make your own LoRA:
https://rentry.org/59xed3
https://civitai.com/articles/91/how-to-correctly-obtain-images-for-a-d ataset
https://civitai.com/articles/3522/vaIstrixs-crash-course-guide-to-lora -and-lycoris-training
https://civitai.com/articles/4100/how-to-create-a-style-lycoris-for-po nyxl-on-low-vram-locally
>Prompt tips for new users:
Grab an image you like the style of from the desired model's civitai's site and drop it onto the PNG info tab in SD, you can then click send to text to image to reuse its generation info for your own images. Change the seed so you do not get just the same image. Use tags from e621 and natural language.
For the love of CelestAI, DO NOT FEED THE TROLLS
>Previous Thread
>>43171882
Showing all 493 replies.
>>
you unironically have to learn art to generate good slop
>what are negative spaces and why are they important
>what makes some perspectives "pop" and why it's a stupid idea to abuse 5-point perspectives in every gen
>how to form interesting compositions
>the importance of values
>color theory (you know your color wheels, right?)
>>
>>
>>
>>
>>
>>43223173
>you unironically have to learn art to generate good slop
No it's so much bigger than this, you have to learn a whole different language to communicate with an alien, an alien who understand english. no code, no art terms or anything like this, we have created an alien intelligence capable of doing tasks impossible to humans.You have to express your desires and translate them into english, and understand how to connect the ideas and concepts, even artists said in private this surpasses everything they know. They know they're cooked because an AI artist will understand this alien language, where an artist will echo what he learned in school or in tutorials for the past decades. See it more like Archimedes discovering quantum physics.
>>
File: Golly.jpg (172.6 KB)
To all the traditional artists and drawfags ITT: stop seething, hating and start steering this train.
AI isn't "stealing" it's a new medium that runs on your work, and right now the pony fandom has one of the best shots at shaping it into something we're all proud of. Purplesmart, PDV6/7, the various NoobAI/Zoinks forks are literally built on the ocean of MLP art that's been posted here, on Derpibooru, DA, etc. for 15+ years. Your lines, your styles, your coloring, your anatomy fixes they're already in the weights.
You already made it possible.
Instead of crying about it mocking the bad gens, trolling AI artists, bullying and hating, why not help make it better? Contribute and own it too, because we are in a community.
Why collaborate? Because models are the future here. A well-made character or artist-style LoRA on PonyXL/Purplesmart lets anyone generate consistent, high-quality versions of your OC, your obscure background pony, or your personal artstyle in seconds while still letting you prompt new scenes, angles, and expressions that would take forever to draw manually.
Training is accessible like I use Kohya_ss, AI Toolkit, or even one-click trainers on Civitai/Tensor.Art. You don't need a 4090 rent on RunPod. Or collab with an AI artist. Use tags like artist:yourname, style:yourstyle properly. Dataset of 20-100 high-quality, varied images (different poses, lighting, expressions, no heavy compression) beats 500 low-effort ones.
The "unethical" cope is bullshit. Every artist studies and learns from others. Human artists copy the show, masters, other artists, trace (rarely with permission), "borrow" styles. AI just does it at scale with math (diffusion + transformers + LoRA adapters). The output quality depends on the data quality and prompt engineering. If you feed it your best work and help curate datasets/LoRAs, you get outputs that respect proportions, clean lines, good coloring, and MLP soul instead of melted garbage.
Imagine looking at a killer art and saying "Yeah, I trained the LoRA / contributed to it for that style.This isn't AIslop, We did this together." That's the biggest contribution you can make to the fandom. This tech is here. Fandoms that embrace it (good data, good LoRAs, good prompting) will have endless high-quality art of their characters. Ones that seethe won't.
Here's my idea for the artists who aren't afraid but choose to be based kings joining the AI revolution, post your style refs, share dataset tips, drop your Derpi tag, Help train a personal artist or OC LoRA and release it. Help Purplesmart and PDV7 get better pony understanding. Turn the "AI slop" complaints into "This is the best pony art we've ever had, because we guided it. It's our baby, it's got our DNA in it" We're all ponies here. Cooperate, collaborate, create. The model improves when you help improve the training. Let's make something legendary and this is the teachings of MLPFIM in action.
>>
>>
>>
>>
File: 1772822325388749.png (2.5 MB)
Cleaned up an old one. Was a bit tempted to make Sunburst calm down enough to stand on fouractually fivelegs but that would've been too much effort and also hard to bake in so this'll have to do.
Forgot to fix a detail.
>>
>>43224320
Natural language prompting. This kind of shit:
>A peaceful woodland scene with a towering, gnarled oak tree in the center, its sprawling limbs reaching skyward. The tree is encircled by dense ferns and wildflowers, with a shallow creek in the foreground mirroring the lush surroundings. A young woman in a flowing dress stands on a mossy boulder to the right, admiring the ancient tree. The afternoon light filters through the leaves, creating golden patterns across the forest floor.
>>
>>
File: Gemini_Generated_Image_m9rs9mm9rs9mm9rs.jpg (2.1 MB)
>>43222663
>>43223608
How do i summen a sucubuss?HOLY SEX btw
>>43222626
Kek
>>
File: Gemini_Generated_Image_r27kxzr27kxzr27k.jpg (2.1 MB)
>>43224969MOMMYYYYYY
>>
File: Gemini_Generated_Image_opm932opm932opm9.jpg (2.3 MB)
>>43224972
Special toll Pipette
>>
>>
>>43224974
>>43224972
>>43224969
New lore just dropped. The liquid is urine. And she's a drug testing nurse.
>"Pee pee in zee bottle" "Now please"
>>
>>
>>
>>
>>
>>43224969
>>43224972
>>43224974
It's impressive that Gemini can do so much, considering that the model is censored to hell and back.
>>
File: now-please.jpg (1.8 MB)
>>43225124
>>43226069
>NOW please.
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: Screenshot_20260508-064441~2.png (877.0 KB)
Working on some animated illustrations to go with a fic I'm working on, and a recorded soundtrack to go with it as well.
There's a shitload of good ways to quickly get good pone gens and images, are there any known to be able to animate still pones / characters like picrel without getting shitty or inconsistent with the art style?
I think I could isolate each part of each image like the crowd here and animate only that, overlay on still background and do the same for other moving figures like Anon on (and off) stage. It's already easy animating the environmental elements with more traditional normie gen software like blowing trees, rustling curtains, moving or dynamic lights, clouds, etcetera
>TLDR are there any good services or software that could possibly be used animating show-accurate pone stills (for example, making the crowd in picrel have heads moving and looking around, or blinking)
>>
>>
>>
File: Gemini_Generated_Image_q6aq9eq6aq9eq6aq.jpg (1.7 MB)
>>43228100
The diagnosis is autism
The prescription is happy pills and neetbux
>>
File: Gemini_Generated_Image_6t38wc6t38wc6t38.jpg (1.8 MB)
>>43229069
The test results have come back and unfortunately (You) will never be a woman
There are happy pills for that, though
>>
File: Gemini_Generated_Image_vg9xs5vg9xs5vg9x.jpg (1.6 MB)
>>43229073
Nursette base image
>>
>>
>>
>>
>>
>>
>>
>>43229073
>>43229575
The chart should have been of a mare vagina.
>>
>>
>>
>>
>>
>>43231371
I guess you have the ones that are currently being licensed by Hasbro? I have Trixie and Starlight and they're really good in terms of how they're held together. With OlyFactory plushies I'm always worried of their manes getting detached.
They have recently released Celestia and Luna, though they're more on a pricey side, but as it is with chink stuff, they will get cheaper.
>>
>>
>>
>>43231415
>I guess you have the ones that are currently being licensed by Hasbro?
Nah they're really old. I think it's either Famosa or Play by Play. AJ has pigtails and isn't wearing a hat. The label doesn't say who made it unfortunately.
>With OlyFactory plushies I'm always worried of their manes getting detached.
I just checked mine and It looks like the stitching holding AJ's hair attached to her head is also coming loose now. :( And I haven't even touched it in ages (I just keep them on the dresser in my bedroom).
>>
>>
File: a.jpg (1.3 MB)
made some porn.
https://files.catbox.moe/980il4.png
https://files.catbox.moe/q50eec.png
https://files.catbox.moe/3sivbh.png
https://files.catbox.moe/go704a.png
https://files.catbox.moe/24s697.png
https://files.catbox.moe/q5rj9v.png
https://files.catbox.moe/7p825p.png
https://files.catbox.moe/tsgtq9.png
https://files.catbox.moe/x98tja.png
https://files.catbox.moe/t4cie6.png
>>
File: aa.jpg (1.2 MB)
https://files.catbox.moe/qy0k85.png
https://files.catbox.moe/ezfay2.png
https://files.catbox.moe/dueunx.png
https://files.catbox.moe/pzuuaa.png
https://files.catbox.moe/c6x32y.png
https://files.catbox.moe/kwtrnd.png
https://files.catbox.moe/acn8u7.png
https://files.catbox.moe/6qauq7.png
https://files.catbox.moe/b3v6yk.png
https://files.catbox.moe/jgwebb.png
>>
File: aaa.jpg (1.9 MB)
https://files.catbox.moe/vsu4v0.png
https://files.catbox.moe/m5xlpq.png
https://files.catbox.moe/y28zhc.png
https://files.catbox.moe/5nrxly.png
https://files.catbox.moe/y5lyth.png
https://files.catbox.moe/3eorrn.png
https://files.catbox.moe/ftmlnv.png
https://files.catbox.moe/i4eeah.png
https://files.catbox.moe/490d7t.png
https://files.catbox.moe/qn1d44.png
>>
>>
>>43231504
>>43231508
>>43231515
wasted on human
>>
File: aaaa.jpg (361.6 KB)
https://files.catbox.moe/ksi0iz.png
https://files.catbox.moe/swn3u4.png
https://files.catbox.moe/kxeze8.png
https://files.catbox.moe/1pcmqh.png
https://files.catbox.moe/qj6g5b.png
https://files.catbox.moe/dzao98.png
https://files.catbox.moe/4jgqna.png
https://files.catbox.moe/fd69g4.png
https://files.catbox.moe/lo5n3c.png
https://files.catbox.moe/y6dvp9.png
>>
>>
File: Gemini_Generated_Image_7wv4q47wv4q47wv4.jpg (2.2 MB)
>>43230768
These aren't perfect but I think they're pretty cool
>>
File: Gemini_Generated_Image_9412t49412t49412.jpg (2.2 MB)
>>43231691
>>
File: Gemini_Generated_Image_df7f1odf7f1odf7f.jpg (2.5 MB)
>>43231698
>>
File: Gemini_Generated_Image_ka46acka46acka46.jpg (2.1 MB)
>>43231703
>>
File: Gemini_Generated_Image_ej62pdej62pdej62.jpg (1.6 MB)
>>43231707
Crystal guard Pipette base image
>>
File: Gemini_Generated_Image_s74mv1s74mv1s74m.jpg (1.7 MB)
>>43230185
Are you sure, Anon? Remember, they're suppositories.
>>
File: Gemini_Generated_Image_5fm3on5fm3on5fm3.jpg (1.7 MB)
>>43231717I did NOT ask for the text "Daily Dose". IT KNOWS
>>
File: Gemini_Generated_Image_rlc6oerlc6oerlc6.jpg (1.8 MB)
>>43230718
The problem is that it does it wrong. I think it only understands human pussy
>>
>>
>>
>>
>>43231717
>>43231725
Thank you doc
>>
>>
>>43231504
>>43231508
>>43231515
>>43231531
Some of these are really cool, I like the sketchy style. Good variety of poses and poners too.Thanks for giving us something other than this homogenous Pipette spam to look at
>>
>>
>>
>>
>>
>>
File: Gemini_Generated_Image_1ce2hk1ce2hk1ce2.jpg (2.0 MB)
>>43234140
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: Gemini_Generated_Image_w8zjemw8zjemw8zj.jpg (2.0 MB)
>>43234716
Cuddles
>>43234727
Noice
>>
File: Gemini_Generated_Image_3qis1z3qis1z3qis.jpg (2.1 MB)
>>43236291
>>
File: Gemini_Generated_Image_m2ekulm2ekulm2ek.jpg (2.1 MB)
>>43236293
>>
>>
>>
>>
>>
Superb gens, as usual!
>>43231717
>>43231725Please tell me you specifically asked for this poster, and that the AI isn't trolling you...
>>
>>
>>
>>
>>
File: 1776926597510146.png (1.4 MB)
>>43237489
Like this?
>>
File: Gemini_Generated_Image_wtzgw1wtzgw1wtzg.jpg (2.0 MB)
3D printed
>>43237660Kek yeah I gave it the image and told it to put it in the background. Would be amazing if AI could troll you that well though
>>
File: Gemini_Generated_Image_zbv8q1zbv8q1zbv8.jpg (1.8 MB)
>>43238448
>>
File: Gemini_Generated_Image_b5ns97b5ns97b5ns.jpg (1.7 MB)
>>43238451
>>
How do I make a character make a heart sign over their pussy? Specifically YiffyMix, but other models based on Pony are fine too. I tried "pussy heart" (the e621 tag for it) but it does nothing no matter how much I put emphasis on it, just make hearts appear in the image.
>>
>>
>>43238700
>>43238469
desu I read that as a vulvar winking. thats what I see when I see his first comment.
>>
>>43238700
>>43238757
No, I mean using her hands to make a little heart gesture over her pussy. I'm talking about anthro abominations, obviously.
>>
>>
>>43238778
>>43238795
Don't forget to show the thread your finished work.
>>
>>43238448
>>43238451
>>43238452
Impressing that the AI manage to generate something so coherent!
Even the 3D printer is plausible. What a time to be alive!Kek, you got me worried for a second. Good trolling then!
>>
>>
>>43238960
>>43237934
NTA but I need fresh content
>>
>>
>>
File: generate_image__replace_all_characters_202605120146.jpg (2.1 MB)
>>43239290
Don't worry, I'm sure I have a fat equestria girl in here somewhere you'll like!
>>
>>43238778
Use LoRAs for things like this.
https://civitai.red/models/2561402/heart-hands-on-pussy
>>
>>43237849
>>43237856
Super cute
>>
>>43238448
>>43238451
>>43238452
This gives me an idea. Pipette as a white marble statue, greeting all visitors with a warm smile.
>>
>>
File: 1750120980110536.png (2.5 MB)
>>43239660
Remacri
https://civitai.com/models/147759/remacri
>>43238960
>>43238984
Brown Kirin to the rescue?
>>
>>
>>43237934
>>43239753
Bros, I love kirins so much...
>>
File: ComfyUI_01325_.png (862.9 KB)
I've started doing some small experiments about the viability of training a large scale pony based finetune of Anima, since I'm tired of local pony gens having to deal with the prompt adherence and anatomy issues of Ponyv6 and SDXL in general.
Results so far have been promising, it adheres to pony anatomy reasonably well, but even though I included a few other non-mane six characters in the small dataset I used, it struggled to adopt them properly (pic related). It also has little to no nsfw capability yet because I didn't include any in the dataset for simplicity's sake. I didn't want to mess around with content tags if I wasn't sure it was going to even be worth it.
>>
File: ComfyUI_01309_.png (630.4 KB)
>>43240246
One of the better mane six images from the test samples.
>>
>>43240246
IMO, even if the narrative seems to shift against PDXL, the score modifiers (when done correctly) are a much better implementation of quality tags than "masterpiece, best quality, very aesthetic" nonsense.
>>
File: 1754780335959350.png (513.2 KB)
>>43240252
Reminds me of StableMond with the "flat colors" tag.
Other Illustrious finetunes with furry datasets can get very similar results. Is base Anima better than those?
>>
File: ComfyUI_01317_.png (758.8 KB)
>>43240267
I never had a problem with the tagging scheme. It's just starting to show its age, and considering most users ended up using it for generic furry and anime shit and v7 was DOA I realized it was unlikely we'd get another model with a significant amount of pony data again.
>>
>>43240272
The main advantage it has is the ability to use hybrid captioning. So you can describe the minutia of the image using natural language and then tack on tags at the end to dial it in and the model doesn't freak out. It also can do text rendering.
>>
>>43240278
Hell, even PDXLv7 got worse at pony, because Astralite was more concerned with making it an "allround" model (and his cashgrab app, of course) than improving on what made his model big in the first place.
Anyway, how are you training your Anima derivative? Are you just merging other models together or training a new dataset (doesn't that require like a dozen of H200s running for months?)
>>
>>43240391
The final goal is a full finetune yes. But this first test was just a regular lora built using a small dataset I scraped together of about 3000 images. About 1000 of which are regular anime images so that the training doesn't outright collapse the model. Another 1000 being general OC images so the model learns general pony anatomy and doesn't collapse into make mane six recolors and the remaining being about 100 images for each character.
As for captioning, I scrapped all the images with their derpibooru tags intact and then used a local instance of Joycaption to inject natural language captions into the tag files.
My next task is to fix the character adherence and slowly introduce more and more characters. Once I'm confident it can reproduce both original characters as well as pre-established characters without concept bleeding I'm going to graduate to a full finetune. I'm hoping that since Anima is already an illustration focused model with basic knowledge about ponies, I won't need to rent a bunch of GPUs for months. Part of the cost of Ponyv6 was that SDXL was a much more general model that had a bunch of stuff the use case didn't want or need, like photoreal and "regular art". Plus Anima is stupidly easy to train. Unlike SDXL or even Illustrious and NoobAI, I still have yet to encounter any friction when it comes to simple stuff like style or character loras. It really does just work.
>>
>>43240391
>(and his cashgrab app, of course)
excuse me but 90% of users use the free version, the 10% are who make this possible and maintained.
But speaking of cash grab, how's your art pack scams going Patachu? What will you be rigging next?
>>
File: Gemini_Generated_Image_p5lvdwp5lvdwp5lv.jpg (1.9 MB)
>>43239555 (checked)
As a marble statue
Btw if you haven't seen it, I did something with marble a while back:
https://tantabus.ai/68636
>>
File: Gemini_Generated_Image_1ep01v1ep01v1ep0.jpg (2.4 MB)
>>43240496
>>
File: Gemini_Generated_Image_em6uiyem6uiyem6u.jpg (1.9 MB)
>>43240500
>>
>>43240501
>>43240500
>>43240496
Adorable as ever. And you're right, I missed that.
>>
File: 1762349060277851.png (2.8 MB)
>>43240429
Just don't overdo it on pony only since there's a lot of stuff in anime, furry, and other stuff that's cool to add to ponies.
>>
>>
>>
>>43240530
I know. That's why a large portion of the dataset is going to be anchoring images, I also have humanized and anthro art in order to further separate the pony concepts so nothing bleed into it that isn't supposed to. to make sure the model doesn't collapse and get ruined.
>>
>>
>>
>>
>>
>>
>>
>>
File: HHNXDTFaQAAJW4E.jpg (170.1 KB)
>>43241039
all the good ones stopped bothering posting here due to the same reason. Dalle 2022 thread here.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: 00018-The717-2026-05-14.jpg (1.2 MB)
>>43241039
No, I just made my own thread for self-containment because otherwise I'd be spamming these threads bc I gen a lot
be nice
>>
>>43242229
Based contributing genchad
>>43241039
Fail complaining leech
>>
>>
File: 00023-The717-2026-05-14.jpg (1.3 MB)
>>43242246
And she's a very cute one!
>>
>>
>>
>>
>>
File: Gemini_Generated_Image_xr1wh0xr1wh0xr1w.jpg (3.2 MB)
>>43241659
The best kind of flood
>>
File: Gemini_Generated_Image_ujixd9ujixd9ujix.jpg (2.6 MB)
>>43242403
>>
>>
>>
>>
>>
>>
>>
>>43242421
Yes, I typically WhiteCaliberFox's or Bendy_and_Boney MLP anthro loras.
I use Equestria Girls and Humanized as additional tags while removing anything animalistic for human gens.
When making regular ponies rather than anthros, I specify with the no_humans, horse, feral, and hooves tags
>>
>>
>>
>>43242051
>>43242178
Seth shares a few AI artists in the Drawfriend posts.I've seen tyto4tme4l and itsrealtorumi in them a bunch. Celjewnia just shits out so much admittedly high quality slop she eclipses everyone else.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: 100737.jpg (456.4 KB)
>>43243334
Only when it's porn. If you're trying to make a quality SFW image the standards should be much higher.
AI is really only good for porn though. I have yet to see anyone here make even a simple image like this with AI. The technology is fine for cooming, but worthless for creating art with any actual merit, and I say this as someone who gens every day (coom, obviously).
>>
>>
>>
File: Gemini_Generated_Image_14rz2l14rz2l14rz.png (1.7 MB)
>>43243758
>>
File: Gemini_Generated_Image_jhcoksjhcoksjhco.png (1.7 MB)
>>43243760
>>
File: Gemini_Generated_Image_wzwvz8wzwvz8wzwv.png (1.5 MB)
>>43243762
One moar
>>
>>
>>
>>
>>
>>
File: 1770226824230991.png (2.3 MB)
>>43244039
Figuring out how much work doing another Kirinventure would be with my current tools and standards.
The answer is quite a lot.
>>
>>
>>43242776
>She
>>43243334
>Tee hee standards whats that?
>Btw join my patreon owo
>>
File: Gemini_Generated_Image_b3ik9ab3ik9ab3ik.jpg (2.6 MB)
>>43242883
The horror, the horror!I tried doing something with Lily Valley lying on the ground but the model was too dumb
>>43243758
>>43243760
>>43243762
>>43243806
Noice
Isn't it great how flexible Gemini is?
Do they have names/stories?
>>43244018
Very cool!
>>43244166
Another Kirinventure would be neat, but I understand if it's a pain to get something you really likeDaily Pipette posting involves compromises
>>43242912
>>43244434
One day Celunia will do a pussy reveal and you guys are gonna look so silly
>>
File: Gemini_Generated_Image_tn199ntn199ntn19.jpg (2.7 MB)
>>43244454
Oh hey, there are mane six clones now too!
>>
>>
File: Gemini_Generated_Image_8806e88806e88806.png (1.5 MB)
>>43244454
Yep, luv me Gemini.
>names/stories
I could talk about my mares all day, but I'll be brief.
>the Zeeb
Zalika, a morally dubious occultist and adventurer coming from a rich family. She seeks out all kinds of arcane knowledge and power.
>The fancy mare
Scarlet Petal, a noble restaurateur and conneisseur of unusual and interesting food. She is kind of a slut.
>the hippogriff
Ocean Twirl, an adventurer mercenary griffmare. She's kind of a charismatic psycho.
>the dragoness
The Winter Lady. Think of her as a not!buddhist demigoddess of nihilism and rejecting desire to numb oneself to pain. Very tsundere.
>Picrel
Silk Road. A hustler, a merchant, a thief and a fence mare. She's in it for the money.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: Nyxiepone.jpg (642.4 KB)
>>43244434
I did join it. Her OC ponies are cute. If she turns out to be a troon im not there for her HV anyway.
>>
>>
>>
File: Gemini_Generated_Image_qmey4gqmey4gqmey.jpg (2.1 MB)
>>43245883
Changeling? What's that?
>>43244514
>Scarlet Petal, [...]. She is kind of a slut.
Kek
>>
File: Gemini_Generated_Image_rd5giprd5giprd5g.jpg (2.4 MB)
>>43246337
>>
>>
>>
>>
>>
File: vaudeville ponka.jpg (3.8 MB)
>Every Zigger Had a Lady But Me (1897)
https://files.catbox.moe/dks9bm.flac
>Hesitate Mr. Zigger, Hesitate (1897)
https://files.catbox.moe/p1ul2d.flac
>When a Coon Sits In the Presidential Chair (1898)
https://files.catbox.moe/kqu1sr.flac
>There's a Zigger in the Moon (1897)
https://files.catbox.moe/vhts36.flac
>I'd Rather Be a Zigger Than a Poor White Mare (1894)
https://files.catbox.moe/ox4v8h.flac
>>
>>
File: 3730274.gif (208.1 KB)
>>43246506
>>
>>
>>
>>
File: 1761330722117236.png (1.2 MB)
>>43247093
The "trick" was her sneaking pineapple in all your food for the past week. And now she is about to reap the rewards of her effort.
>>
>>
>>
I have 15 free uses of the Tsubaki.2 model left on PixAI but for the life of me I can't seem to gen ANYTHING reasonable looking with it. I tried both tag based prompting and natural language, and the prompt helper (which just rewrote my natural language in slightly different natural language). Any tips? Otherwise, if you have any requests (pony obviously, but anthro or humanized versions is fine too) feel free to make them and I'll try to gen it for you, otherwise they're just going to expire anyway.
>>
>>
>>43246337
>>43246340
N-Never mind, I hear my laundry calling, gotta go!
>>43246506
Neat!
What did you use? Cause it's really clean!
Also, the voice is good but not perfect. I wonder, if you make it sing one of the show song, what would be the result?
>>43247454
Have you tried to ask an LLM to make the prompt for you? (like chat GPT, Grok or whatever?)
>>
>>
>>
>>
>>
>>
>>43247719
Here's the tool I used: https://colab.research.google.com/github/ShiromiyaG/RVC-AI-Cover-Maker -UI/blob/main/assets/RVCAICoverMake rUI.ipynb#scrollTo=sDo4MkcLcN5l
And the voice models I used: https://huggingface.co/therealvul/RVCv2/tree/main
There's a local version, but I'd recommend trying the google colab version first so you don't have to fear downloading anything shady to your PC. All you have to do is drag and drop whatever song you want to convert, the RVC voice model you want to use for the conversion, and it'll do the rest for you. Don't mess around with the RVC settings at all except to change the pitch depending on the gender of the original audio vocals (+12 for male-to-female, and -12 for female-to-male).
Only drawback is that mixing potential is limited due to lack of extensive options (could only adjust reverb and volume of the vocals, instrumental, and backing vocals respectively), so if you really want to have a clean-sounding song, you're gonna have to find a song with vocals that synergize with the specific RVC model you pair it with.
Here's a short example of a really good AI conversion IMO:
https://files.catbox.moe/0etyeh.mp3
>Also, the voice is good but not perfect. I wonder, if you make it sing one of the show song, what would be the result?
Unless you want to make, say, Twilight sing a Pinkie song, that would kinda be unnecessary since the original already suffices lol.
>>
>>43247851
Thanks for the details!
RVC improved since I last used it in the pppv
But it was one year (maybe two?) ago, so there is that.
>Useless
Well, no, it allows you to control the synthetic voice. The AI won't just say "nope, the singer is already good", it will still try to swap.
And that way you can compare with the original and evaluate the accuracy of the AI.
But thanks for posting resources, if I have some time, I will try my silly idea myself.
>>
File: Gemini_Generated_Image_dm7yj8dm7yj8dm7y.jpg (2.3 MB)
BOOM *ch*
BOOM BOOM *ch*
>>43247129
Freckles...
>>
File: Gemini_Generated_Image_647g2z647g2z647g.jpg (2.2 MB)
>>43247881
>>
File: Gemini_Generated_Image_ngv0b2ngv0b2ngv0.jpg (2.3 MB)
>>43247884
>>
File: Gemini_Generated_Image_oert3loert3loert.jpg (2.1 MB)
>>43247887
>>
File: Gemini_Generated_Image_xu6c6jxu6c6jxu6c.jpg (2.2 MB)
>>43247890
>>
>>43247859
Oh, I see where you're coming from. Nah, I gotchu.
Here's RVC Rainbow's singing model applied to her own song I'll Fly from S5.
I'll Fly OG (ground truth): https://files.catbox.moe/7bzigf.flac
Since it's basically her own voice, I disabled the search feature ratio setting(or index file) by dialling it to zero (if you don't know what that is, it basically converts the base vocals' accent to correspond with the RVC model. Since the accent of the base audio perfectly corresponds with the RVC model, there's no need to apply it here).
RVC Dash I'll Fly (SFR 0) - https://files.catbox.moe/2fsoq7.flac
Now here's the same but with the search feature ratio - or accent - turned up to 100:
RVC Dash I'll Fly (SFR 100) - https://files.catbox.moe/0m4058.flac
The biggest thing I noticed is that the alveolar sounds that RVC Dash makes are notably lighter than the OG song.
>>
>>43247937
>>43247859
For a more apples-to-apples comparison, I converted the raw vocal stem from "This Strange World" using Twilight's RVC singing model.
This Strange World vocal stem OG (ground truth): https://files.catbox.moe/iq1wp0.wav
This Strange World RVC Twilight (SFR 0): https://files.catbox.moe/4304pr.flac
>>
>>43247937
Damn, you rock!
It's really nice to be able to compare the three.
My hearing may not be perfect of course, but the OG still sounds rouder, more full.
The is still really good especially when you don't compare with the OG.
I feel the AI a bit more "harsh" and dynamic, more like a stage performance.
Most of the time, I can ear a difference with AI and OG, the voice is not the same (close, very close, but not the same).
I don't think I would be able to tell which one blindfolded tho.
As for SFR0 and 100, they are indeed different, but again, I wouldn't be able to tell which is one (or better).
Thanks for taking the time to appease my curiosity.
I really should lurk the /pppv/ again, it's have been so long.
Now that I have resources (new job, yay!), I don't have time unfortunately...
Life is ironic sometimes...
Now I wonder what vocoded voices would give...
Like "Harder better faster stronger" from Daft Punk...
>>
File: 1766981953096816.png (1.2 MB)
>>43247799
>>43247839
>>
>>43247956
Should have refreshed before posting.
Yeah, here the difference is more audible.
Especially when she take a "slutty" voice ("But I don't have much time with them, got to learn all that I can")
The AI almost sound like it's mixed with Trixie sometime.
But again, I don't know if I could tell so in a blind test.
I probably would find the voice strange, but add music, sfx and images, and I think I would be fooled.
>>
>>
>>
File: 1749490025008795.png (1.1 MB)
>>43248243
>"Couldn'tcha just squeeze these cheeks all day?"
>>
>>
>>
>>
>>43248263
>>43248272
>>43248275
Kirin really are just too cute, danke!
>>
>>43247960
Peaceful, lovely!
>>43248364
this one super cute!
>>
>>
>>
>>
>>
>>43248945
One of the Saddle Arabian visitors.
>>
>>
>>
>>
>>
>>
>>
>>43231504
>>43231508
>>43231515
>>43231531
Sweet. Ponified EqG next?
>>
File: Gemini_Generated_Image_1uv3bu1uv3bu1uv3.jpg (2.0 MB)
I've been sitting on this idea for a while because I figured it would be censored and I'd have to figure out a workaround
Decided to just try it today and Nano Banana Pro did it with basically no complaint. I only got one rejection in 50+ gens
>>
File: Gemini_Generated_Image_hpsopphpsopphpso.jpg (2.0 MB)
>>43249500
>>
File: Gemini_Generated_Image_rczp1krczp1krczp.jpg (2.0 MB)
>>43249502
>>
>>
>>
>>
>>
>>
>>
File: Gemini_Generated_Image_dm5ev2dm5ev2dm5e.jpg (2.1 MB)
Ponytail
>>43249826
The fun part is that it's not even my GPU
>>
>>
>>
>>
>>
>>
>>
>>43252862
Honestly, I don't see any improvements anatomical wise in any of the more recent models compared to PDXLv6. Illustrious, NoobXL or Anima, they all make the same stupid errors just like the original base model they're based on does. All that differs is the default style and support for e621 tags and the like. But YiffyMix makes just as much conjoined fetus ponies as PDXL does.
>>
>>
>>
File: ComfyUI_01481_.png (957.1 KB)
I retrained my model with a better tagged dataset, the anatomy seems to be generally better however the more obscure characters I included still aren't taking, I think I might just need more images and I'll have to start included nsfw in the dataset to do that. Over all through having a flexible pony generator feels nice. I tried included more complicated looking actions and poses to show off Anima's capabilities.
https://files.catbox.moe/rvcwfu.png
https://files.catbox.moe/ggsm3f.png
>>
File: ComfyUI_01496_.png (808.2 KB)
>>43253794
A few more since the retarded captcha system kept saying my post was invalid.
>>
File: Gemini_Generated_Image_4zhg3i4zhg3i4zhg.jpg (2.3 MB)
>>43251518
PiratepetteI don't think paying for a subscription counts as piracy though
>>
File: Gemini_Generated_Image_3a12n13a12n13a12.jpg (2.4 MB)
>>43253860
>>
File: Gemini_Generated_Image_9t2hm29t2hm29t2h.jpg (2.7 MB)
>>43253862
>>
>>43253860
>>43253862
>>43253864
Cute pirate mare.
>spoiler
Oh. I didn't know that.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: Gemini_Generated_Image_q8dibhq8dibhq8di.jpg (2.1 MB)
>>43254727
This one was very tricky, and I'm not super happy with the result. It was difficult to get the AI to generate a good image of a necklace with a potion gem, and then even harder to get it to put it on her properly. I think the problem is that the necklaces don't really make sense and wouldn't go around a pony's neck properly, so it really wants to do something like...
>>
File: Gemini_Generated_Image_e7ndfge7ndfge7nd.jpg (2.1 MB)
>>43256022
...this
>>
>>
File: Gemini_Generated_Image_5z6hwh5z6hwh5z6h.jpg (2.0 MB)
>>43256024
>>
File: Gemini_Generated_Image_1pk5ij1pk5ij1pk5.jpg (2.0 MB)
>>43256025
>>
>>43256022
>>43256023
The first one is looking fine. I really like the gleaming effect that it creates. And the second one is funny if nothing else.
>>
>>
>>
>>
File: Gemini_Generated_Image_p3e5vap3e5vap3e5.jpg (2.3 MB)
>>43256858
Tropical fruit drink
>>
File: Gemini_Generated_Image_nkbgm0nkbgm0nkbg.jpg (2.2 MB)
>>43258100
>>
File: Gemini_Generated_Image_tvnoeutvnoeutvno.jpg (2.3 MB)
>>43258103
>>
>>
>>
>>43223157
Hey, I appreciate all the links being dropped in the OP. I recently bought a potent computer and would love to start generating short ai videos of Ponies kissing the pov.
I am tech illiterate tho, so other than opening the links and skimming through what's inside I don't understand what I am reading, the high amount of subfolders and command lines that appear mid GitHub explanation like --something something install something is very much frightening. I assume comfyhub is the easiest to get my feet wet? I wish there was something like a YouTube video to follow, it is easier to learn by seeing someone do it than reading.
Alas, just venting.
>>
>>43258538
OP here, find a content creator that you like on youtube and follow their vids. But to start off with I think stability matrix is the best front end installer for a beginner. It will download everything for you. Then you just need to decide if you want to jump straight into the chaos and go comfyui or a forge fork. I think Forge Neo would be ideal since I've been doing it for years and only use comfyui for workflows that require it. For videos I would even recommend finding a patreon sub that has everything in one stop. But no one wants to hear that. But I am telling you if you want a simple one click installer with all the bells and whistles installed then its your best bet. SECourses or whatever is breddy gud.
>>
>>43258538
>I wish there was something like a YouTube video to follow
Protip: youtube has a search feature.
Basically, ComfyUI (assume that's what you mean by "comfyhub") or Forge is your interface to provide parameters and run model(s). Choose whichever one you like, but I'm not sure if Forge supports video gen. Both of these can be installed by Stability Matrix, which is a kind of downloader/launcher for everything related to local generation. If you can use git and python virtualenvs, then you don't really need it - readme in Forge/Comfy repo should have all instructions. If you have a very recent GPU then you might need to install manually anyway, in order to download the correct version of some python libs.
Get models and LoRAs from civitai or huggingface and put them in the directory designated for models (again, check readme).
Good luck. You might want to start with basic image gen before jumping on video gen.
>>
>>
>>
File: Screenshot 2026-05-22 212929.jpg (181.6 KB)
I don't know if this celjewnia reads this thread, but you did good. Ignore the faggots in the comments who can't figure out how to turn on filters on Derpibooru.
>>
>>
>>
>>
>>43258103
When you realize AI draw sitting pones better than the animators themselves...
>>43259275
Oh shit, it genially hurt to see what we could have if hasjew took the right decisions!
It's glorious, and, unfortunately, just a dram...
>>
>>
>>
>>
>>
File: skateboarding-no-sound.webm (3.8 MB)
THE GAME HAS CHANGED
Version with sound: https://tantabuscdn.net/img/view/2026/5/23/80183.webm
>>
>>
>>
>>43260400
Pipette pro skeatboarding! Yay!
But seriously, damn, that's impressive. How does a video like this cost?
>>43260439
Great, now AI can solve anti AI captcha.
>>
>>43260812
>Great, now AI can solve anti AI captcha.
Just like with mail spam, in the future 98% of our energy consumption will be going to AI scrapers, bots and algorithms made to break AI detection. And people will still refuse to see that this world is infected with a cancercalled capitalism
>>
>>
>>
>>
>>43259275
I'm imagining if you fall down bottomless pits in this theoretical game they do a skyward sword sort of thing where it fades out and fades back with a wonderbolt telling you to watch your step before flying off.
>>
>>
File: OIG4.asjry8zC.jpg (145.4 KB)
>>43261102
On a slightly related note, I wonder if anyone ever modded ponies into Palworld. Seems like a good fit
>>43259343
Reminds me of how I felt about a block game before Minecraft came out, or a serviceable Magic the Gathering before Hearthstone (and MtG Arena following). And we know those games did pretty well. You cannot undersell the value of giving the people what they want.
>>
>>
File: file.png (124.6 KB)
I plan on fully sharing all of the Olsen-Banden films soon, though I'm not sure which board I'll do so. For whatever reason no one had ever dumped these online with English subtitles.
I'm not sure on which board I'll share them, I've been steadily uploading them to internet archive, however it's going to take a while since they have a very slow upload speed.
Would /t/ accept links to the internet archive, or would I get banned for not seeding myself?
>>
>>43259275
>>43260400
Sweet!
Can you generate an image (I guess video is still expansive) of a low poly (PS1, N64, Saturn) game like these? Hey, I have weird tastes, I know.
>>43261173
That would be cute!
>>43261377
I don't think it was ever required to seed the link your share on /t/.
Heck, some bastard there are even anti seed...
>>
>>
File: Gemini_Generated_Image_hkbx51hkbx51hkbx.jpg (1.8 MB)
>>43260812
>How does a video like this cost?
I don't know dollar amounts, but to make that I genned 3 different videos:
https://u.pone.rs/fmcakfad.mp4
https://u.pone.rs/swetuxqj.mp4
https://u.pone.rs/fszcgcdq.mp4
Each used about 25% of one of my 5-hourly limits on the Pro (~$20/month) plan. I made the final video by stitching together good bits from the first two gens.
>>43261319https://tantabus.ai/64816
>>43261398
>Can you generate an image (I guess video is still expansive) of a low poly (PS1, N64, Saturn) game like these? Hey, I have weird tastes, I know.
There is actually already a low-poly skateboarding Pipettesee the spoiler above, but I made more for (You).
>>43261411
You get 5-hourly limits and also a weekly limit. The weekly limit feels very generous to me, it seems like you would need to fill up a lot of 5-hourly limits to hit it. A video gen takes up about 25% of a 5-hourly limit on the ~$20/month plan (like I said above), but it also seems like it's very slightly over 25% - I couldn't get a full 4 gens starting from 0 when I tried. It looks like there's an ~$8/month plan that has half the limit of the ~$20/month plan, as well as two tiers above at ~$100/month and ~$200/month, which give 5x and 20x the limit of the ~$20/month plan respectively.
>>
File: Gemini_Generated_Image_tvktsltvktsltvkt.jpg (1.9 MB)
>>43261590
>>
File: Gemini_Generated_Image_siswb9siswb9sisw.jpg (1.7 MB)
>>43261594
>>
File: Gemini_Generated_Image_lhn7y4lhn7y4lhn7.jpg (1.6 MB)
>>43261597
I don't know what the Pony-pump is and I'm not sure I want to find out
>>
>>
>>43261590
>>43261594
>>43261597
>>43261599
Sweet, thank you for these gens!
And thank you for all the details too.
>>
>>
>>43262128
It's a Danish comedy series spanning from 13 films from 1968 to 1981. The main characters are low tier criminals who at best manage to get coins from a coin booth, buth they constantly try and always fail to pull a big score. It usually starts with the group's leader, Egon Olsen getting arrested while his partners flee, and then getting released from prison. In prison he usually comes up with another get rich scheme. Their heists are elaborate. Since they're bunch of losers, they don't ever fight since they wouldn't stand a chance against actual brutes. That's why Egon's schemes rely on using the environment to their advantage, timing things right, and creative use of everyday items (like a baloon, a toy car and a piece of string to bypass security alarm if I remember it right, or whipped cream to disable it).
Despite their efforts, things pretty much always go wrong, either bad luck, third parties involved or their own incompetence. By the end of the film, Egon usually ends up getting arrested again.
Egon is the shortes of the bunch, smart, but has short fuse and can get dramatic by how things don't go his way.
Benny is their driver, the tall one, the best he can manage is picking some locks. Always cheerful and happy to help.
Kjeld is the nervous, chubby one, carrying a doctor's back with tools for Egon. Kind of acts like a decoy. Victim of bybossy wife, Yvonne. Yvonne's strict demands for exactly how they will spend their non-existent millions frequently derail Egon's carefully laid plans as she has a tendency to mix up their loot.
The first film has some 'early installment weirdness' as it features a slightly different setting than the rest of the series. The boys are associated with some hookers who aid them when needed, Kjeld has more children. Since the second film he's only got one teen son, Borge who is smarter than he looks like and takes advantage over others with ease.
I don't know how they have pulled this off, but all of the films are entertaining to watch and feature a story that you want to follow and see its conclusion.
>>
>>43262167
For whatever reason, they made a 14th movie in 1998 which features the main actors as senile men, and it's simply sad to watch. I won't even bother with it, but Kjeld's actor died the same year it premiered, Egon's in 2004 and Benny's in 2009. I don't know what was their mindset while filming this. It's sad because it has good concepts for the heists, but it would be much better as some sort of reboot if anything. Having your main actor suffer from dementia in plot, the other on a wheelchair throughout the entire movie isn't comedy, it's a scary portrayal of getting old. Yvonne's absence is also very notable in this film, as her actress had also died somewhere in mid 1980's I think.
Anyway I'm missing two films as of now, the 11th and the 12th. I'm not sure whether to hold out with sharing the collection now or not, because I'm importing these DVDs from Denmark and it might take a week or two before it's fully complete. That said, it's already partially up on the internet archive. Once I send the 9th movie, I think I'll wait for the 10th and 11th to arrive, and then upload them alongside with the 13th.
First movie if anyone's interested.
https://archive.org/details/the-olsen-gang-olsen-banden-1968
>>
>>
File: 363bba0f_no_sound.webm (1.3 MB)
Oh my god you guys, this is so amazing
With sound: https://tantabus.ai/images/80332
More:
https://tantabus.ai/images/80331
https://tantabus.ai/images/80330
https://tantabus.ai/images/80329
I upgraded to the $100/month tier btw. It seems to only 2.5x my video gen limit but I'll take it.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: a6b5af8c_no_sound.webm (3.7 MB)
These videos are melting my heart
With sound: https://tantabus.ai/images/80420
>>43262544
I wouldn't normally be this spendy, but this is a big deal to me
>>
File: 198d1823_no_sound.webm (3.6 MB)
>>43264296
With sound: https://tantabus.ai/images/80419
>>
File: 7addd64f_no_sound.webm (3.6 MB)
>>43264298
With sound: https://tantabus.ai/images/80418
>>
>>
>>43262353
Damn, that's cute!
>>43264296
>>43264298
>>43264300
And these are cute too!
Now that you can do video, the cuteness will go in overdrive!
$100 is still some money, but I guess you can do a lot aside from video for you.
On the other hand, you are very not far from making shorts. Only need a small scenario, and you are good to go!
>>
>>
>>
File: c30a0fa8_no_sound.webm (3.8 MB)
>>43264549
This one would have been perfect if it wasn't for the strange walk at the start
With sound: https://u.pone.rs/hynqprbz.mp4
>>43264588
>Now that you can do video, the cuteness will go in overdrive!
Absolutely! It's gonna be so good. I love the close-ups it does.
>$100 is still some money, but I guess you can do a lot aside from video for you.
Probably. I haven't actually tried any image gens yet so I don't know how it affects the limits for those.
>On the other hand, you are very not far from making shorts. Only need a small scenario, and you are good to go!
Yeah, I have been thinking about that. I don't know if I will actually do it, but I think it's amazing that we have the power to do that so easily now. I was thinking that at this point, we basically have everything we need for fan-made MLP episodes. I guess we already did, but it just got a lot easier.
>>
File: bee4e598_no_sound.webm (3.7 MB)
>>43264549
>>43265364
This is something I stitched together from a few different clips
With sound: https://u.pone.rs/unuatdrj.webm
Raw gens:
https://u.pone.rs/whfphfgv.mp4
https://u.pone.rs/lhmrquza.mp4
https://u.pone.rs/ekqjlxdv.mp4
>>
>>
>>43265364
>https://u.pone.rs/hynqprbz.mp4
The sound is pretty solid too. I haven't expected that.
>>
>>
>>
File: 1772271924819906.png (2.1 MB)
Boop.
The hard part wasn't getting a pony in power armor. Nor was it genning a mutiecorn.
No, it was the raised hoof. That would turn into a hand no matter what. This is the best I could do.
>>
>>
>>
File: f764878c_no_sound.webm (3.7 MB)
THIS IS SO COOL
With sound: https://tantabus.ai/images/80485
>>43265376
Kek true
>>43265693
Yeah it's cool
The sound makes a big difference
>>43265775
Lmao I didn't notice that
I like that video a lot more now
>>
File: a66e140a_no_sound.webm (3.7 MB)
>>43266075
With sound: https://tantabus.ai/images/80484
>>
File: 1779817835283308.png (2.0 MB)
>>43265824
>No, it was the raised hoof. That would turn into a hand no matter what.
We desperately need a pony-focused model like ponyXL but for Illustrious and Noob.
>This is the best I could do
In cases like this, just edit it yourself. It's not that hard. Run it through inpaint at low denoising strength, around 0.1–0.2
>>
>>
>>
>>
>>43266103
I don't want to spend too many gens on itor take the time to clean them up lmao. I made these three for you though:
https://u.pone.rs/dsnxzqkf.mp4
https://u.pone.rs/gnbrvkkq.mp4
https://u.pone.rs/ojsmpyed.mp4
Not perfect but I think there are some nice moments there
>>
>>
>>43266075
>>43266077
Kek, these are even better in movement!
Just missing some "hoa!" from Pipette, and Jurasic Park music, and you are good!
>>
>>
File: OIG..Y15TTSilDnHrs0bVR4k.jpg (142.7 KB)
>>43266579
There are a lot of considerations, like would it be effective as a LoRA, or should the entire model be finetuned on an MLP dataset? According to LTX's devs, it needs to be trained on clips whose length is (a number divisible by 8) + 1 frames. What can we do to get accurately rendered cutie marks? How many ponies should the dataset include? Do we exclude Twilicorn? The crystal tree castle? The School of Friendship? Do we want to include detailed descriptions of the characters so that OCs can be generated using detailed descriptions?
One approach I've thought about is to put two copies of each clip into the dataset. One of them would name the character, the way we'd actually prompt it: "Twilight Sparkle levitates a book." The other one would describe the character and what levitation looks like: "With a magical sound, a purple magical aura appears around the horn of a unicorn who has a straight purple mane with a magenta stripe and blunt bangs, whose cutie mark looks like a six-pointed magenta star surrounded by five small white stars. A purple magical aura surrounds a book as she levitates it."
>>
>>
>>
>>
>>
>>
>>43266139
Apparently there's also a request limit that's been set to 3 recently: https://discuss.huggingface.co/t/free-account-zerogpu-quota-issue/1751 80/9
Guess I'll just crank up the number of inference steps to 8 if I only get to make 3 videos a day anyway.
>>
>>
>>
>>
>>
File: b82ac336_no_sound.webm (1.7 MB)
Ball
With sound: https://tantabus.ai/images/80639
Full mp4: https://u.pone.rs/ctdezuoa.mp4
>>43267456
I checked it out and I get 10,000 credits per month. An Omni video gen costs 30 credits, so in theory it gives me ~11 more video gens per day.As well as, of course, the opportunity to blow through all my credits in a day and then maybe give Google even more money.It seems like you can get about 25 gens from the app in an average day without ultimately exceeding your weekly limit, so I'll probably use the app more, but getting extra gens from Flow is very nice, so thanks for pointing that out.
>>43266570
>>43267146
>>43267679
>>43268237
I gave both
>music in the style of Jurassic Park
and
>Jurassic Park music
a try. They were both allowed and it seems like they do the same thing, which is Jurassic Park-esque music that presumably is not directly lifted from a movie.
Some gens (there's a lot of errors in these):
https://u.pone.rs/chuupmgs.mp4
https://u.pone.rs/vrzkfqya.mp4
https://u.pone.rs/clovdyue.mp4
https://u.pone.rs/mmmdiqob.mp4
https://u.pone.rs/xnvykcps.mp4
https://u.pone.rs/nwdufuhk.mp4
https://u.pone.rs/qyuhfhmw.mp4
https://u.pone.rs/pxarpqdc.mp4
>>
>>
>>
>>43268320
No problem pal. By the way, the best use-case for google omni is both image-to-video and video-to-video. If you're still relying on text prompts for your outputs, a better workflow would probably be to generate image with gpt/gemini with desired subject matter -> use google omni and specify desired video output or download random video -> replace subject with favourite pone
I can't use video-to-video with omni because google is cucking UK users atm, but you could give it a shot
>>
>>43268342
>I can't use video-to-video with omni because google is cucking UK users atmSame.I'm prompting by adding Pipette image(s) and telling it how to make the video. Video-to-video would be very useful as long as it didn't degrade quality too much.
>>
>>
>>
>>
>>
>>
>>
>>
File: f47da273_no_sound.webm (3.6 MB)
Bongos
With sound: https://tantabus.ai/images/80699
>>43268640
>>43269082
>>43269294
In fairness, it's not just us this time:
https://support.google.com/gemini/thread/435692494?hl=en&msgid=4359120 83
>Video to video (edits) are not available in the following countries/regions
>EEA, India, Switzerland, the United Kingdom, and some US states.
>>43269682
Wings were the main issue when I was genning images. Now with videos, horns are a common problem too. I figured genning an earth pony would be easy!
>>43269808
Go Anon!
>>
>>
>>43259275
We have pony game at home:
Pony game at home:
https://www.youtube.com/watch?v=cFVM53u22Mo
>>
>>
>>
>>
>>
>>
>>
File: 9d80d375fed980744abf276d95a53a81.jpg (76.1 KB)
I'm sure it's a long shot but I'd like to request Anon (don't mind either human or pone) dangling from Rain Shine like he's just barely holding on enough to bury his face in her mane but not land on his feet if he fell, if that makes sense. Possibly with another pony or kirin looking on slightly embarrassed by him but I don't want to push it.
>>
>>
>>
File: 1767166320074303.png (734.3 KB)
>>43271507
Not the stance you wanted but have this. Background is DIY.
>>
>>
>>
>>
FASHION
With sound: https://tantabus.ai/images/80775Almost broke the chain today. Getting good video gens is way harder than getting good image gensAnd I accidentally posted these to Tantabus out of order FUCK but at least I got trips
>>
File: 9bd2a4fa_no_sound.webm (1.7 MB)
>>43272554
Oops, didn't add no-sound webm
>>
File: 86049116_no_sound.webm (2.2 MB)
>>43272558
With sound: https://tantabus.ai/images/80777
>>
>>
>>43272558
>>43272560
The dresses are nice.But it is slightly ruined by the human clothing in the background.
>>
>>43272086
>>43272117
This is super cute, thank you very much!
>>
>>
>>43272931
>spoiler
Kek yeah, I originally wasn't going to post them because of that, but it got to the end of the day and I didn't have anything I felt was better. I figure it's OK because bipeds exist in the FiM world.
>>
>>43272558
>>43272560
Damn, that's mindblowing!
The AI manage to "understand" what a dress should look like on an equine.
The purple dress in the second video especially, where it "understand" that there is a tail under there, and it needs to accommodate.
What a time to be alive!
>>
>>
>>
>>
File: bacc3e98_no_sound.webm (3.5 MB)
Explorign
With sound: https://tantabus.ai/images/80867
>>43273597
>What a time to be alive!
It sure is.
>>43273719
Kek
>>43273937
Local models will catch up eventually. Until then, just pay $100/month and only generate SFW stuff. What's the problem? ^:)
>>
>>
>>43274351
I showed it https://tantabus.ai/images/73919 and https://tantabus.ai/images/73917, for which the prompt was something like
>Create a square image showing this character walking up some stairs in a world reminiscent of the works of M. C. Escher. Don't include people. Maintain the realistic style.
The final video is the result of me taking parts of many different video gens and combining them. The prompt for those video gens was something like
>Create a video based on these images. Show the pony exploring the world. Include a close-up of the pony's face. The pony shouldn't speak or make any noises. Don't include music.
>>
>>
File: 1669259042326487.png (159.1 KB)
>>43259275
>mfw twitterfags repost the sloppa I made to the tune of 200k views and hundreds of comments
https://xcancel.com/LDrawzL/status/2058587422162919604#m
well, now. wouldn't dare to take pride in it or anything since it's just ai sloppa, but it does feel breddy good seeing your own creation out in the open like that. Even saw a few of my dbz gens posted on facebook, too, kek
>>
File: HJJmjm0WwAEDlne.jpg (265.7 KB)
>>43274551
Well, you certainly didn't create the video or a game, but you did materialize a bit of what people want, and that's nice.
So, yeah, it's sloppa, but it have the same utility as "the pones" (the video mocking SimS1 but with ponies): showing what could be possible and make people dream.
So, yeah, there is still that.
>>
>>
>>
>>
>>
>>
File: f7cf63ee_no_sound.webm (3.5 MB)
A relaxing day at the RAINBOW waterfall
With sound: https://tantabus.ai/images/80975
>>43274415
>You have the power to make your own Pipette episodes anon!
Kek I totally do, but for now I think I want to stick to more basic stuff.Also I wouldn't need a writefag's help, I could just ask AI
>>43275313
I'd been wanting to see that too, so I tried it out. I wanted to let the AI do whatever it felt was right, so I just gave it a basic
>Create a video based on this image.
prompt. I'm pretty pleased with the results:
https://u.pone.rs/ybxsskrx.mp4
https://u.pone.rs/cujwgudd.mp4
https://u.pone.rs/azbwvlvz.mp4
https://u.pone.rs/vhxnedps.mp4
https://u.pone.rs/epjkqzgd.mp4
https://u.pone.rs/cuvtezoa.mp4
>>
>>
>>43276172
"They say knowledge is power. I prefer to think of it as a deliciously wicked advantage."
Dayum
Also, for scenario, you really want to either drive really close the AI, or ask for a writefag's advice. AI only gives run off the mill scenarios without real depth and rarely interesting.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>43278000
Seems https://huggingface.co/spaces/r3gm/wan2-2-fp8da-aoti-preview-2 doesn't give the error, but every failed run counts towards your maximum daily runs too. How annoying.
>>
File: 802e5c67_no_sound.webm (3.7 MB)
Riding on a RAINBOW owl
With sound: https://tantabus.ai/images/81061
>>43276325
>>One immediately pulls away her clothes
That one would have been SO GOOD if it didn't add the second pair of wings
>I think my favorite is the 'Where's the fun in that?' one.
I can see the appeal. It lacks errors and has a nice close-up shot.
>>43276328
>"They say knowledge is power. I prefer to think of it as a deliciously wicked advantage."
>Dayum
The voice is pure sex
>AI only gives run off the mill scenarios without real depth and rarely interesting.
True tbdesu
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
File: 1760462230453367.png (2.5 MB)
"I'll have the one on the top shelf. Yes, the blue box."https://u.pone.rs/fdkaause.png without censorship