File: c9697914-cd3b-472b-a56a-4(...).png (2.56 MB, 2304x1344)

2.56 MB PNG

>>3093930

File: 00019-4153734477.jpg (1.2 MB, 1280x1728)

1.2 MB JPG

Glad to see a new thread, ty for caring about us. Trying to keep the new images coming. /Y/ is a pretty slow board overall, just trying to do my part!

File: 00056-2213507399.png (2.65 MB, 1728x1296)

2.65 MB PNG

2nd of all, I would like to add that you can rent GPU processing over the web, cheaper than paying any website to generate with you.

For example, looking at Vast.ai, you can rent an RTX 4090 (24gigs vram) for 0.36 cents per hour. You can rent for as little as one hour at a time, and only pay as you go.

Now of course, you would need to have a good size SSD on your home PC, like 200 gigs free. And some regular ram matching the GPU you are renting, 32 gigs is pretty standard these days. Not to mention a good highspeed connection. But if you can manage that, lets do a little math:

>NovalAI seems to charge 0.036 cents an image, claiming 384 "basic" images for $14.

>Conservatively, lets say you can make 30 images an hour renting @ 0.36 cents per hour

>That same $14 gets you 35 hours with a good GPU

>That equals 1050 images at a much better quality by renting

The "better quality" I am talking about is including upscaling and ADetailer fix.

I also bet you can make more than 30 images an hour. But you have to included any lag time for loading programs on your PC, not to mention actually writing out prompts.

As an example, I can make 1 size 1280*1728 (upscaled and ADetailer fixed) image on an illustrious checkpoint with 16 gigs vram and 32 gigs system ram in just over 60 secs.

Just food for thought for anyone interested.

File: 00042-4051407124.jpg (1.35 MB, 1728x1280)

1.35 MB JPG

Mind you, I am not trying to knock sites like NovalAI or CivitAIRed, but you have to stretch your money as much as you can.

If you can download these programs and checkpoints on your PC, but don't have a Graphics card to match, you are better off learning to rent the GPU.

File: 00016-521884526.jpg (1.5 MB, 2048x1536)

1.5 MB JPG

In Other News:

I have some recommendations for Anima Based generation.

I finally found a few ways to make it better:

>https://civitai.red/models/2851133/mindshift

This checkpoint is the best male focused one I have seen yet. Works pretty decent out of the box so far as style goes. It also follows prompts pretty well.

>https://civitai.red/models/2840614/huggingcute-or-anima

Another good male focused checkpoint, but as the webpage states, aimed more at cute magna style twinks, (cough*cough*) I mean femboys.

Both of these checkpoints make pretty good bulges and penises without any odd lumps the way base anima can. They both also work with body types other than "bara", unlike:

>https://civitai.red/models/2708104/homosimile-anima

As I just mentioned, a lot of the trained checkpoints coming out are losing ground in prompt adherence. So even if they have a better style, the checkpoint stops understanding some important nuance in prompting. And the better, more complex prompt adherence is all Anima really has over illustrious SDXL type anyway.

File: 00013-3344733808.jpg (1.12 MB, 1240x1824)

1.12 MB JPG

Continuing on Anima related suggestions:

>https://civitai.red/models/2673444/anima-uncensored-llmclip-new-int8?modelVersionId=3001834

This model can be used in place of the Qwen3 based text encoder. In my testing it actually does work better for NSFW than the original.

>https://civitai.red/models/599757/velvets-mythic-fantasy-styles-or-flux-pony-illustrious-zit-anima-krea2?modelVersionId=3226360

VelvetS released an Anima version of the "Smooth Lines" style Lora. On the dark side, but really good, and keeps prompt adherence.

>https://civitai.red/models/2633578/animavolumetric-glow-fantasy-style-illustration-aesthetics?modelVersionId=2976629

This style Lora can come up with some interesting pictures. It does need these prompts added: volumetric glow, soft cinematic lighting, dreamlike atmosphere, luminous materials

>https://civitai.red/models/2564225/style-greater-gpt-image-2?modelVersionId=2998634

This is a good all around "realistic" anime style, with good details and prompt adherence.

>https://civitai.red/models/2458426/anima?modelVersionId=3126581

The lab behind anima itself did release this checkpoint on top of the base. It does have a better style, but still keep putting "masterpiece, best quality, score_7, ... " in positive prompt and "worst quality, low quality, score_1, score_2, score_3, artist name, blurry, jpeg artifacts, chromatic aberration" in negative prompt.

The anima aesthetic 1.1 goes very well with the mentioned style loras, but they work well on the checkpoints mentioned here as well: >>3093937

File: orig_00050_.png (1.22 MB, 1536x1024)

1.22 MB PNG

https://files.catbox.moe/14cd2h.mp4

Testing minimax h3, I guess I suck at prompting and it understood that I wanted them to talk like sims

>>3093934
Love his pits.

File: 00004-2791886356.png (1.78 MB, 1296x1728)

1.78 MB PNG

>>3093985
lol, it is a nice attempt. I have a lot of issues getting penis right in Krea 2, and it seems H3 has issues as well. Oh well.

File: 00111-3054029649.jpg (1.12 MB, 1728x1280)

1.12 MB JPG

>>3094017
I have come to appreciate a good armpit thanks to image generation.

File: 00151-2135611721.jpg (1.1 MB, 1728x1280)

1.1 MB JPG

Among other interesting things...

File: 00107-1375300561.png (3.3 MB, 1536x1536)

3.3 MB PNG

...and some even stranger things...

File: 00048-1110531687.jpg (612 KB, 1536x1152)

612 KB JPG

Still trying to get strange things like demonically possessed dildos to work though

File: 3.0__masterpiece, best qu(...).png (1.54 MB, 1024x1024)

1.54 MB PNG

mmm monster men

File: kirito_bodysuit_elf_.png (1.56 MB, 896x1152)

1.56 MB PNG

>>3093943
loving how dark this image came out, also thanks for the suggestions

File: 1768470179_9851_12906.png (1.49 MB, 1024x1024)

1.49 MB PNG

Not sure if this the proper place for this kind of semi-realistic style, but where else should I post it?

File: 00234-295396460.jpg (1.24 MB, 1728x1280)

1.24 MB JPG

>>3094134

Semi-realistic stuff is OK.

I had tried some of the checkpoints that do realistic all the way. The characters come out looking like cosplayers though, so I never really got into it.

File: 00092-595657105.jpg (1.01 MB, 1280x1728)

1.01 MB JPG

>>3094137

Tokito is so fucking cute, I hate the fact he is canonically 14, only reason I haven't done a set of him yet. I hate trying to age them up and hope it passes some censor

File: 00094-2771566590.jpg (1.26 MB, 1280x1728)

1.26 MB JPG

>>3094091
<3

>>3094136
Beautiful
Can you share model and prompt?

File: c5c29698-3c33-47ba-88ca-8(...).png (3.36 MB, 2304x1344)

3.36 MB PNG

Hottest man in all media for me and it's just John Finalfantasy

File: 088cdc44-e437-4447-b957-9(...).png (3.18 MB, 2304x1344)

3.18 MB PNG

>>3094191
washing up a grown ass man

File: 3b53c6d1-8917-4a42-8374-e(...).png (3.29 MB, 2304x1344)

3.29 MB PNG

>>3094192

File: 00053-3657467392.png (2.14 MB, 1728x1296)

2.14 MB PNG

>>3094191
Isn't it great being able to make porn of so many characters that never had enough?

File: 2boys, nsfw, danganronpa,(...).png (1.34 MB, 832x1216)

1.34 MB PNG

>>3094177
All of those were done months ago by first doing a NovelAI 4.5 prompt (see pic), and then putting it through CyberRealisticPony as a reference image. I have the full NAI prompt, but don't remember the details of CRP, other than the guidance scale must be pretty low to keep the anime aesthetic. At full strenght it just completely changes the hairstyle and face shape.

Prompt: 2boys, nsfw, danganronpa, komaeda nagito, large penis, deepthroat, ejaculation, cum overflow, cumdrip, bukkake, cum bubble, komaeda nagito giving a deepthroat, very aesthetic, masterpiece, no text

Undesired Content: lowres, artistic error, film grain, scan artifacts, worst quality, bad quality, jpeg artifacts, very displeasing, chromatic aberration, dithering, halftone, screentone, multiple views, logo, too many watermarks, negative space, blank page

Character 1 Prompt: boy, danganronpa (series), komaeda nagito, facial, cum in nose, cum on hair

Resolution: 832x1216
Seed: 3275975563
Steps: 28
Sampler: k_euler_ancestral (karras)
Prompt Guidance: 5

File: 2boys, nsfw, danganronpa,(...).png (1.42 MB, 1024x1024)

1.42 MB PNG

>>3094135

File: 00007-1011176022.jpg (1.11 MB, 1280x1728)

1.11 MB JPG

>>3094210

Boobs?!?

File: br02.png (2.8 MB, 1008x1792)

2.8 MB PNG

>>3094216

File: br03.png (2.98 MB, 1008x1792)

2.98 MB PNG

>>3094218

File: e524e6bd-68d3-4e5a-a3ea-a(...).png (3.22 MB, 1344x2304)

3.22 MB PNG

>>3094200
It's amazing, and dangerously addictive.

I tried h3 again with kinda better results. https://files.catbox.moe/j3bcf9.mp4

>>3094200
>>3094223
I love being able to finally see anime style characters with body hair.

File: 00129-1696892814.jpg (1.13 MB, 1728x1296)

1.13 MB JPG

>>3094226

For the longest time I didn't care much one way or the other about body hair. But once I started experimenting with it, getting the hair just right, I realized how much it can add to the overall character. It just feels unreal to not have any hair now.

File: 75daff70-ca65-4ca4-a48a-c(...).png (2.43 MB, 1344x2304)

2.43 MB PNG

Just a challenge between bros

>>3094019
god damn thats nice. prompt?

File: 00143-2828091574.jpg (1.08 MB, 1728x1280)

1.08 MB JPG

>>3094236
This is from the same batch. Note it was made using:

>https://civitai.red/models/579632/indigo-furry-mix-xl?modelVersionId=2074658

And not Homosimile, per my usual.

Positive: masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest, very awa, nsfw, male focus, 1boy, teenager, inuyasha, animal ears, dog ears, white hair, yellow eyes, claws, long hair, completely nude, slim body, pectorals, nipple piercing, biceps, male pubic hair, animal penis, red penis, veiny penis, knot, canine sheath, cock piercing, taint piercing, anus, sheath ring piercing, testicle piercing, necklace, precum, precum drip, seductive, serious, looking at viewer, dark, light, abstract background, <lora:yiru:0.85> Smooth_Quality

Negative: facial hair, mature, saliva, worst quality, bad quality, jpeg artifacts, signature, username, Smooth_Negative-neg

Steps: 40, Sampler: Euler SMEA Dy Negative, Schedule type: SGM Uniform, CFG scale: 7, Seed: 2828091574, Size: 1152x864,

Model: indigoFurryMixXL_noobaiEPS7, Model hash: 8f212f8bc2, Denoising strength: 0.45, RNG: CPU,

ADetailer model: fdetailerAdetailerFor_v11.pt, ADetailer confidence: 0.3, ADetailer use separate sampler: True, ADetailer sampler: Euler, ADetailer scheduler: Karras, ADetailer version: 25.3.0,

Hires Module 1: Use same choices, Hires upscale: 1.5, Lora hashes: "yiru: a5c35919b190", TI: "Smooth_Quality, Smooth_Quality", Hires sampler: Euler, Hires upscaler: 8xNMKDSuperscale_150000G, Hires steps: 18, Hires CFG Scale: 7,

Version: neo-2.28

>>3094238
thanks man!

File: 00122-2744157670.jpg (878 KB, 1728x1280)

878 KB JPG

>>3094241

File: 7d02ba9b-09e7-4353-bf1c-c(...).png (2.47 MB, 2304x1344)

2.47 MB PNG

Animated: https://files.catbox.moe/lvb8vm.mp4

File: 00008-2552311327.jpg (1002 KB, 1728x1280)

1002 KB JPG

>>3094252
Kabru is hot. A little to far on the fem side for me personally, but definitely attractive.

>>3094219
>>3094218
>>3094216
I like those although they could clean up once in a while.

>>3094216
>>3094218
>>3094219
Prime stay-at-home husband material, but the bastard has to be clean if I'm going to slave away to support him!

File: ComfyUI_ZIP_video_000004.webm (717 KB, 640x1088)

717 KB WEBM

Not so known YouTube series but I knew I wanted to do this when I saw the original scene

File: 00201-3170694673.png (2.54 MB, 1536x1536)

2.54 MB PNG

Glad to see posters interacting with the thread this week. Good to see other peoples ideas

File: 00034-942739496.png (2.31 MB, 1536x1536)

2.31 MB PNG

Hope everyone is having a good week.

File: ComfyUI_ZIP_video_00000 (2).webm (1.31 MB, 640x1120)

1.31 MB WEBM

Sorry for the spam, having so much fun with this

some of the anima style loras I made:

https://litter.catbox.moe/nabzau5q67a469n4.safetensors

https://litter.catbox.moe/0obw6qg7clhwh972.safetensors

https://litter.catbox.moe/rre8t7ivaxcxq8y4.safetensors

File: 00127-2920650598.jpg (1.17 MB, 1280x1728)

1.17 MB JPG

>>3094312
Do you have any examples? Are these posted elsewhere, like CivitAI or Tensor.art?

File: 00106-2604138794.jpg (1.33 MB, 1280x1728)

1.33 MB JPG

>>3094312

Typically speaking, there are trigger words involved in activating these Loras as well.

>>3094315
I don't use trigger words (always inferior - I tested a bunch of times with anima just to be sure) and no they're not on any website. I also don't feel like making samples, just uploaded some of the ones I made on a whim

File: 00093-3654536625.jpg (1.14 MB, 1280x1728)

1.14 MB JPG

Just a few pictures to keep the board moving

File: 00015-3633079140.jpg (855 KB, 1728x1280)

855 KB JPG

Nothing crazy interesting

File: 00082-104672275.jpg (937 KB, 1728x1280)

937 KB JPG

a few odds and ends

File: 00121-2750526397.jpg (1.38 MB, 1280x2048)

1.38 MB JPG

Hope everyone has a good weekend

File: 00015-265393253.png (1.68 MB, 1296x1728)

1.68 MB PNG

And has the time to generate something interesting!

File: f9c50272-e858-45aa-8a5d-b(...).png (3.32 MB, 1344x2304)

3.32 MB PNG

>>3094358
Ooc, do yo use a Lora for Leon? I struggle to recreate him on anima. Hot picture.

File: 00077-1047984803.png (1.99 MB, 1296x1728)

1.99 MB PNG

>>3094368

Yes, a character Lora was needed for illustrious and Pony.

Try these tags at least, as I don't think a Lora for him exists for Anima:

kuwata leon, danganronpa \(series\), 1boy, ear piercing, blue eyes, short hair, red hair, spiked hair, goatee, white shirt, open jacket, long sleeves

File: IMG_20260906_041016_075.jpg (185 KB, 746x1280)

185 KB JPG

>>3094369
Thanks, a part of the goatee, I liked how accurate it is.

File: VID_20260906_041026_095.webm (3.01 MB, 640x1152)

3.01 MB WEBM

>>3094382
An attempt to animate it (I'm addicted to h3, if anyone want to try it I kinda found a way to try it for free)

File: 00144-336307533.jpg (1.28 MB, 1728x1280)

1.28 MB JPG

>>3094383
Ya know, I tried using Wan 2.2 before, to animate a "start" and "finish" picture set I made in illustrious. It kind of was working, but it was replacing the illustrious art style with its own sort of bland, shinyness.

Did you start by making the Leon character in H3? Or can you take a picture and have someway for the style to stay?

File: 0d86d7fd-efdb-47e5-a404-0(...).png (2.95 MB, 1344x2304)

2.95 MB PNG

>>3094384
It has really good character retention, it's better if you prompt like a lot but I'm lazy and my prompts are rather short and the results are still nice, but I'm still experimenting. There are two modes, one with just a picture or a start and end pictures, and other with a lot of references where I believe you can use video too, but this last one is too hard to use for me. I just use picture to video and start with an Anima picture, like this one. Also there are a bunch of civitai made loras already that makes easier to do certain things.

File: MP4Video_00002.webm (2.59 MB, 640x1152)

2.59 MB WEBM

>>3094388
The result is this (with audio: https://files.catbox.moe/xb4vy2.mp4), you can do 1 to 15 seconds but I feel that the less the easier to control the weirdness.

>>3093930
Say, how good is NAI (novel AI) at editing / inpainting in general?

>>3094384
Do you use Wan locally? Or is there some online uncensored version?

File: 00190-1394523920.png (2.16 MB, 1296x1728)

2.16 MB PNG

>>3094392
Yes, you can download Wan (and H3) models from CivitAI

>https://civitai.red/models/2100616/wan-22-i2v-8-steps-nsfw-fp8-model?modelVersionId=2376670

You need both the "high noise" and "low noise" version, switching from high noise (to start with) to low noise (to finish with).

It takes more Vram to run then I have on my system (I have 16 gigs) so it is slow for me. I haven't experimented much with it at all, but I can confirm that it is functional using Forge Neo.

You can generate uncensored video on CivitAI.Red, but it is far from free. Just at a glance it is saying 800 "buzz" for a 5 second video. 1K buzz is $1 (USD).

Pic unrelated

File: 1014612934074076285_00101_.png (1.4 MB, 936x1320)

1.4 MB PNG

After 20+ hrs of trying to learn this i still cant use Loras right.
But all he links here /y/ where realy usefull, thanks anons

File: UntitledGoblin19.png (1.38 MB, 1216x832)

1.38 MB PNG

Love making these self-indulgent goblins since there's so little art of hot goblin dudes out there. Still hard to get the faces right though, sometimes they're way too old, ugly, bumpy, and wrinkly. Tough to strike that balance where they're still kind of smooth and sexy.

File: 00005-133223520.jpg (1.34 MB, 1536x1536)

1.34 MB JPG

>>3094418
Sexy goblins eh? Never thought about it before. I will have to give it a shot one day. Has potential.

File: 00045-3874129548.jpg (883 KB, 1296x1728)

883 KB JPG

Orcs work pretty well, but I wouldn't call them "sexy". But it is nice to watch them have fun.

File: 00064-1139040180.jpg (1.28 MB, 1728x1296)

1.28 MB JPG

>>3094418

Try prompting for ages. For example, "21 years old" (or "18 years old") in positive, "Mature" ; "Wrinkled Skin" in Negative.

Depending on what censorship you are up against it might not work though.

>>3094424
That might do the trick, not sure how age ranges work for smaller fantasy races but worth a shot. Because when I'm not specific sometimes they just get super old and wrinkly.

File: 00355-3597398187.png (3.11 MB, 1296x1728)

3.11 MB PNG

>>3094425

You will have to report back and tell us how it went. I imagine that the matrix should be smart enough to figure out your intentions though, and I bet it will help.

>>3094414
Damn, Chris became daddy!

File: MP4Video_00001.webm (2.06 MB, 640x1152)

2.06 MB WEBM

With sound https://files.catbox.moe/tpqlg9.mp4

>>3094392
If you or anyone is interested, I use this app on runninhub (there are better ways on the page but I find this easier and I can use it on my phone): https://www.runninghub.ai/ai-detail-mobile/2084642419408113665?outputId=2096059460856582145

The max resolution permitted in standard is 0,7 and Max duration is 15 seconds, but you will need to experiment a bit. The loras i recommend to look on civitai to see which ones suits better what you are trying to do and how to prompt.

You get 100 coins per day that it can be good for one or two videos. If you use this code in your profile you and me will receive 500 coins: akacecsi, (it's my code but every profile has one and you can use them too). I also made in perchance a tool that helps with the prompting, I'm not a pro so probably it's not perfect but has help me to improve the animations so far so I put it here in case it's useful for someone: https://perchance.org/h3-minimax-prompt-generator

File: 00154-3179232684.png (2.21 MB, 1296x1728)

2.21 MB PNG

Just a few posts to keep the board fresh

File: 00011-4041111172.jpg (1 MB, 1296x1728)

1 MB JPG

The short videos are cute

File: 00093-1469689448.jpg (1.08 MB, 1280x1728)

1.08 MB JPG

Still too lazy to try and make them myself

File: 00033-1342142427.jpg (2 MB, 2048x1536)

2 MB JPG

one of these days ill bother

File: 00043-2508537378.jpg (1.12 MB, 1728x1280)

1.12 MB JPG

Until then I appreciate the inspiration

>>3094491

Rex is S-tier twink. I fucking love him.

>>3094492
would love to see this hottie in a high leg thong

File: 00134-1647687238.jpg (1.05 MB, 1536x1536)

1.05 MB JPG

>>3094502

Rex is hot. A-tier for me, but still hot. Another character I haven't done a production run on because of up-aging. Oh well, maybe someday.

File: 00196-2220306268.jpg (1.18 MB, 1728x1280)

1.18 MB JPG

>>3094511

8 days from now ill be doing a Michael Kaiser (from Blue Lock) posting here:

>https://civitai.red/user/gtgiygus/posts

Today is last of Romeo from Lies of P, and tomorrow starts 7 days of Pinocchio (also Lies of P).

Michael is pretty hot, but matrix kept putting blue flowers around the image, taken from his tattoos. Cute but annoying.

File: ee2f36ba-56a9-430f-9260-c(...).png (2.48 MB, 1344x2304)

2.48 MB PNG

>>3094491
Thanks, it's just my new hyper fixation. I tried h3 on my computer and it's mega slow so when I found a place where I could use it easy and quick i got excited.

File: MP4Video_00004.webm (1.81 MB, 640x1152)

1.81 MB WEBM

>>3094525
https://files.catbox.moe/wmxrq3.mp4

File: UntitledGoblin22.png (1.4 MB, 1216x832)

1.4 MB PNG

>>3094426
Fairly decent results with both "young adult" and "21 years old", though with the randomness I do wonder how much it impacts it. But it did stop the super ugly ones from popping up.

>>3094526

His hole and dick look a little weird, but it's impressive how far AI is coming.

File: 00025-1960311254.jpg (1.02 MB, 1536x1536)

1.02 MB JPG

>>3094527

Glad to hear it is helping.

File: 00082-3337246220.jpg (1.17 MB, 1280x1728)

1.17 MB JPG

>>3094530

I personally cannot get dicks to look right on Krea 2. Never tried H3 but they look just like that on Krea. One of the reasons I haven't used Krea much at all.

File: gorr gape 00012-5950.png (3.91 MB, 1728x1296)

3.91 MB PNG

>>3094538
it really takes danbooru / e621 for models to understand actual nssfw

File: 00033-488834457.jpg (1.19 MB, 1728x1280)

1.19 MB JPG

>>3094593

Crippling gape addiction aside, you have a point. It will take awhile for a good Lora to come out and help here.

File: 00252-500840637.jpg (975 KB, 1280x1728)

975 KB JPG

Posting a bit to keep the thread moving.

File: 00124-284103292.png (2.58 MB, 1536x1536)

2.58 MB PNG

I am actually pretty pleased with the engagement this time.

File: 00272-1774115435.jpg (933 KB, 1728x1280)

933 KB JPG

A lot of neat pictures and ideas to look at.

File: 00067-1060589034.png (2.33 MB, 1296x1728)

2.33 MB PNG

Hope everyone is having fun

File: 00016-516547353.png (1.16 MB, 1024x1280)

1.16 MB PNG

I'm bored, when are they gonna release a new technological breakthrough :(

someone use their gpt astra and fable to create something new lol

File: 00066-2909818086.jpg (1.27 MB, 1296x1728)

1.27 MB JPG

>>3094696
Not to "mansplain" or "acktually...", but remember that we are hitting the upper limit on hardware resources of higher end home systems.

File: 00007-487725906.png (2.14 MB, 1296x1728)

2.14 MB PNG

All the latest releases, like Krea 2 and Qwen, expect you to have 24 gigs Vram and 64 system Ram.

File: 00034-2198981054.jpg (916 KB, 1296x1728)

916 KB JPG

Yes, it can function with less, and there are "Turbo" Lora and Checkpoint Loras that can help.

File: 00199-2127058430.jpg (1.37 MB, 1280x1728)

1.37 MB JPG

But the more compact versions definitely get way less creative, and slow generation times steal fun away

File: 00138-497751983.jpg (1.29 MB, 1536x1536)

1.29 MB JPG

SDXL based checkpoints (Pony ; Noob ; illustrious) are still being used because they function fully with lesser systems.

Anima was an attempt at better architecture is a smaller size, but most people feel it is lacking.

File: 6aeba442-179e-41bb-9c8e-0(...).png (2.61 MB, 1344x2292)

2.61 MB PNG

Which sampler do you guys use? I'm using gradient estimation / seeds2 + beta57 lately, I really like seeds2 but it's rather slow.

File: 00071-346949574.jpg (1.26 MB, 1728x1280)

1.26 MB JPG

>>3094721

The choice in sampler/scheduler usually boils down to speed vs accuracy.

The type of image generator you are using also matters. Anima calls for those two specifically, but Krea 2 wants different, simpler ones.

Those newer types you like are slower, but should be more accurate to the prompt. The "RES4LYF" family are gradients between speed and accuracy as well, to give you scaling options for your specific needs.

I personally use Euler SMEA Dy Negative / SGM Uniform for illustrious, which is the matrix type I still use the most. That combo is fast, but also (somewhat) inaccurate.

But I want that inaccuracy, because it leads to surprising, almost imaginative, results. Utilizing the newer ones will give me the same picture 4 times in a row almost, if I don't change the prompt. Euler SMEA DY Negative / SGM Uniform will give almost radically different results, even if it stays pretty true to the prompt.

My work flow is to almost always generate 4- 8 pictures with an identical prompt, 2 - 4 images at 864*1152, then flip the orientation to 1152*864. The more *imaginative* combo will really surprise me sometimes, and then I pick my favorites from 4-8 pictures of that particularly scenario.

However, sometimes it really can't follow the prompt and gives me garbage, so I switch to Extended Reverse-Time SDE / Simple. My (not exhaustive) testing showed me that Simple can be pretty accurate and keep down odd noise type artifacts (like 6 fingers, or white background), while the Extended Reverse-Time SDE keeps a bit of imagination.

Really, there is nothing else you can do but experiment to see what you like.

File: 00025-1385762532.png (2.82 MB, 1536x1536)

2.82 MB PNG

>>3094721

It seems to me you always like to post, idk how to put it, "normal" images? No magic, or monsters, or abstraction. Just a good likeness of the characters with a big dick doing gay shit.

I like to do weird, magic like stuff a lot, so I prefer some oddness in the matrix.

>>3094819
This is just what I wanted, thanks. Euler a + SGM uniform is one that I used till recently almost always, and Euler SMEA Dy was one I used a lot when I played with illustrious, not the negative one tho. I have to experiment more with these. I've never though of my gens as normal, they are just things that turn me on that I usually don't find material lf. But I suppose if we compare them with yours they are relatively normal, yours are really creative.

File: 1014612934074076285_00043_.png (1.43 MB, 1128x1128)

1.43 MB PNG

>>3094397
After 400 images, this shit is too hard. I give up. Im too dumb or loras and anima is too hard to figure out.
Ill try SDxl later,

File: 00018-2726466565.jpg (1.64 MB, 2048x1536)

1.64 MB JPG

>>3094822

I see nothing wrong with that picture. The style seems true to the anime show, they have 5 fingers, characters don't seem blended.

What is your specific issue? What program or website are you using to generate? What is it you are trying to accomplish?

You can always message me at

>https://civitai.red/user/gtgiygus

If you rather.

File: asdsadsadfew.png (1.01 MB, 1718x1536)

1.01 MB PNG

idk if it's still worth doing local because almost everything is shit, either anima or noobai
skill issue with prompting probably but still

File: 00130-2367360556.jpg (1.2 MB, 1728x1280)

1.2 MB JPG

>>3094863
Again, I am happy to answer any questions. That picture definitely seems to have prompting issues.

File: 00150-2236365085.png (2.45 MB, 1296x1728)

2.45 MB PNG

I have been generating local since SD 1.5. Took a lot of practice to learn how to prompt, and I always have something new to learn. But it definitely isn't dead.

File: 00202-3674786506.jpg (1.3 MB, 1280x1728)

1.3 MB JPG

>>3094875
Is that Shinji?

File: ComfyUI_11227_.png (1.9 MB, 1280x1280)

1.9 MB PNG

>>3094863
What checkpoint are you using? I haven't had anything come out that messy since StableDiffusion first dropped. I imagine it is a prompting issue as gtg said.

>>3094877
Glad you're still keeping the thread alive. I got a new computer and haven't set up comfy yet, but I will soon so I'll have some new stuff to share.

>>3094881
This shit was Anima

>>3094882
But it's just pure anima output (just upscaled) because I don't use detailers lol

File: 00041-2575794385.png (2.21 MB, 1296x1728)

2.21 MB PNG

>>3094882
>>3094883

You can upload one of the pictures you had issues with to

>https://catbox.moe/

post the link here, and we can read the generation information. This can lead to actual advice on how do to better.

Alternatively, tell us the Anima checkpoint you used, your user interface (Forge? Comfy?), settings (sampling/schedule ; resolution, CFG ; step count (both starting and hires)) and we can see what went wrong where.

>>3094891
Thx but I think I got the problem - too vague prompting I was trying not to emulate artists or characters lol and did just the basic descriptions

Tried the literally same prompt with homosimile and it's way better lol

https://files.catbox.moe/4qmdx6.png

Now to learn detailers lol

File: 00030-1447882741.jpg (1.2 MB, 1536x1536)

1.2 MB JPG

>>3094892

Just some advise:

New prompt:

Positive: very awa, masterpiece, best quality, absurdres, newest, perfect quality, detailed background, intricate details, male focus, explicit, 2boys, penis envy, penis size difference, looking at penis, toned male, 19 year old boy with white hair and blue eyes and small penis, 25 year old man with black hair and green eyes, foreskin, phimosis, completely nude, shower, tiles, water

Negative prompt: child, shota, lowres, worst quality, bad quality, bad anatomy, signature, username, logo, watermark, artist name, old, bad hands,

It very much wanted to push this into shota territory, was kind of hard to make them both adult looking.

illustrious based checkpoints DO NOT use "score_10, score_9, score_8". Those scores are from Pony based checkpoints. Anima copied them just because people were used to them, but you still don't really need them for Anima, only Pony.

Although all checkpoints are different, you really shouldn't need quality tags at both the beginning and end. Or if you do, you should only need a few at the start, and a few more at the end.

You are upscaling HUGLY high. Start with a resolution of 1024*1024, or 864*1152 for portrait or 1152*864 for landscape. Stick to no more than 2x larger. If that isn't big enough for you, it is better to do another run using the large finished picture under "Ultimate SD Upscale (forge, not sure what comfy has). Point is, do it in 2 rounds, not one.

illustrious type checkpoints want clipskip set to 1, not 2. Setting it to 2 was a pony thing. Also, CFG could be higher. 4 is a low.

File: 00042-3030501629.jpg (1.25 MB, 1728x1296)

1.25 MB JPG

>>3094892
Forgot to mention, you can set the output to be a lossless jpeg file instead of PNG. Will make the pictures use way less memory without losing any information or quality

Also, your denoise strength for upscaling can be higher for illustrious/pony models. I use 0.45

Detailing models:

>https://huggingface.co/Anzhc/Anzhcs_YOLOs/blob/main/Anzhc%20Face%20seg%20640%20v2%20y8n.pt

This is a good one for face only. Is made to not redraw the hair. But it needs to see eyes and hair in order to pick something up. Unlike the basic ones faceyolo, that redo the whole head almost.

>https://huggingface.co/ashllay/YOLO_Models/blob/main/segm/CockAndBallYolo8x.pt

This is a good penis one. For best results you need to write a separate positive and negative prompt. For example:

Positive: very awa, masterpiece, best quality, absurdres, newest, perfect quality, detailed background, intricate details, male focus, explicit, foreskin, phimosis, completely nude,

Negative prompt: lowres, worst quality, bad quality, bad anatomy, signature, username, logo, watermark, artist name, old, bad hands,

Notice there is no mention of the characters or background. Just the penis.

>https://civitai.red/models/1228695/fdetailer-adetailer-for-furries?modelVersionId=1384450

This one is trained to pick up faces, cocks, anus, and a few other things. Sometimes it works great, sometimes it redraws an anus to be a little face with eyes and hair. You can't really use a separate prompt because it is picking up on many things at once. Can help though here and there, but you are better of starting with the 2 dedicated ones.

Suggested Detailer settings to change from default:

Detection set to 0.3-0.5. If it gives you trouble set it higher.

Inpainting used Separate steps, and the default 28 steps is good. Use separate sampler and scheduler. I use Euler/Karras, but other simple combos work as well.

>>3094896
thanks for models and spoonfeeding my retard ass

>>3094895
as for upscaling, I do upscale through 4x-UltraSharp, then downscale it back to 2x and then redraw it with detailer lora. Right track imo but still playing around

File: 00145-1408155455.jpg (1.24 MB, 1728x1280)

1.24 MB JPG

>>3094897
There is so much to learn about this stuff. So many models, so many checkpoints, so many steps, outputs, inpainting, etc. etc. I feel almost obligated to pass on knowledge, as imperfect as it may be.

>as for upscaling, I do upscale through 4x-UltraSharp, then downscale it back to 2x and then redraw it with detailer lora. Right track imo but still playing around

We may be talking past each other here, but I don't think you are on the right track exactly.

Upscaling with 4x-UltraSharp is fine

(I would recommend 8x Superscale: https://civitai.com/models/292030/8xnmkd-superscale150000g?modelVersionId=328284),

but there should be no reason to shrink the picture.

In fact, you are saying "downscale it back to 2x", when you should mean "0.5", which would be half. It would seem, and the giant picture you catboxed would be evidence of, you are doubling the size twice in a row, when really it should just be once. The detailing step should be before the upscale step, as the upscale should only make it bigger, not really add anything.

If you do in fact want your pictures to be that huge, it can be done it two steps, but that usually steals detail, not add them. Yes, there are some was using Ultimate SD Upscale to get that big without making it worse. But, again, unless you want the huge size, is really not necessary and takes a lot of time.

Try just generating the one image at 1024*1024, run your Detailers, then upscale 1.5x or 2x with a higher denoising strength, like 0.4 or 0.45, and see if you like the results.

>>3094898
noice pic, saved for future fap

meanwhile me: fucking with detailer because my pics look like shit after swapping upscaling and detailing

File: HS2xDetail_00001.jpg (641 KB, 720x1280)

641 KB JPG

>>3094901

nvm, it was seed fuckery because I usually try to do same seeds because I thought that it'd lead to more coherent changes across regenerations lol

File: HS2xDetail_00003.jpg (632 KB, 720x1280)

632 KB JPG

>>3094902
fuck wrong pic

File: 00020-2524585030.png (2.56 MB, 1296x1728)

2.56 MB PNG

>>3094902

Keeping the same seed only really helps if you are changing or adding a small part of the prompt, like one single word. You have to keep everything else the same, including all generation technical stuff.

OR

you keep the exact same prompt, but change something technical, like how many steps you used, to see how it changes. This is to get a better handle on how those things, like steps and samplers, effect everything else, and maybe make it better or worse.

File: RayGunn1.png (1.1 MB, 768x1280)

1.1 MB PNG

Ray Gunn from the upcoming netflix movie of the same name. The things I'd do to this man...

File: RayGunn2.png (1.1 MB, 768x1280)

1.1 MB PNG

>>3094925

File: RayGunn3.png (1.12 MB, 768x1280)

1.12 MB PNG

>>3094926

File: RayGunn4.png (1.12 MB, 768x1280)

1.12 MB PNG

>>3094927

File: comfy_video_000 (1).webm (3.77 MB, 768x1376)

3.77 MB WEBM

>>3094942

File: milo_2.jpg (2.37 MB, 1536x2720)

2.37 MB JPG

>>3094950

File: 00061-378570974.jpg (1.21 MB, 1728x1280)

1.21 MB JPG

Just a few pictures to keep the thread fresh

File: 00142-1317540486.jpg (1.19 MB, 1280x1728)

1.19 MB JPG

Nothing too crazy, just odds n ends

File: 00099-3937264146.jpg (910 KB, 1280x1728)

910 KB JPG

Hope everyone is having a good start to the week

File: 00014-367777781.jpg (1009 KB, 1536x1536)

1009 KB JPG

And that everyone can generate something interesting

File: 00009-1000147262.png (2.52 MB, 1728x1296)

2.52 MB PNG

But play carefully together now!

File: HStitchTest_00007.png (3.29 MB, 2720x1536)

3.29 MB PNG

Starting to love even manual inpainting (as in mask an area and prompt until perfection) but segs models do shit so much better still

File: AO_00002.png (815 KB, 1360x768)

815 KB PNG

>>3094992
for context, that pic was based on this. just detailing with homosimile and manual cropping and prompting on "Hashida"'s dick because it was a fucking mess. But now it's fine by me, unlike this shit. Anima's such a shitshow

>>3094896
if i may, asking for spoonfeeding again 'cause I can't wrap my head around managing multiple people's features? Got around a bit with one man but can't figure multiple. Multiple instances of detectors or what?

File: 00199-286457617.png (2.46 MB, 1296x1728)

2.46 MB PNG

>>3094993
Getting the older models (Pony/illustrious) which homosimile is, to reliably generate 2 sperate people can be vary difficult. The following prompt is for the included picture:

Positive; very awa, masterpiece, best quality, absurdres, newest, perfect quality, detailed background, intricate details, male focus, nsfw, 2boys, two 19 year old boys, Caucasian, teenagers, slim body, hug from behind, groping, briefs, detailed bulge, male school uniform, body hair, school classroom, pants pulled down, light hair color, messy haircut, fade haircut, worried, nervous, <lora:cosmic_fantasy_art_illustriousXL:1>

Negative prompt: age difference, size difference, old, solo, facial hair, fat, heavy, hefty, bara, erection, girl, women, lowres, worst quality, bad quality, bad anatomy, signature, username, logo, watermark, artist name, old, bad hands

In this case, I used prompts that only work for 2 different characters. By not giving specific details about those characters, I let the generator figure it out on its own.

Both are similar, but different.

If you don't care that much about what the characters look like, this would work.

File: 00045-1524610645.png (2.6 MB, 1296x1728)

2.6 MB PNG

>>3094995

This example, again used in the illustrious model type Homosimile, uses one generic and one known and well trained character.:

Positive prompt: very awa, masterpiece, best quality, absurdres, newest, perfect quality, detailed background, intricate details, nsfw, male focus, official art, 2boys, age difference, sora \(kingdom hearts\), kingdom hearts with brown hair and blue eyes, fat ugly man with black hair and beard, groping, anal, rough sex, motion lines, motion blur, looking at viewer, seductive, heavy breathing, saliva, open mouth, tongue, fog, light, dark, floating particles, <lora:cosmic_fantasy_art_illustriousXL:0> <lora:NOOB_EPSv1_1_detailer_by_vlnvk_v1_0:1>

Negative prompt: facial hair, child, erection, lowres, worst quality, bad quality, bad anatomy, signature, username, logo, watermark, artist name, old, bad hands

The prompting for very different features and body types, plus the fact that Sora had a decent 1000+ image training data set, made it work.

Notice that the character descriptions are in one sentence, with no " , " separating them. If I prompted to make the fat old man a teenage boy instead, the matrix would have started blending them together, probably ending up with a clone of Sora fucking himself. Which is hot, i guess, but not what I was going for.

File: 00000-3951131310.jpg (1.22 MB, 1536x1216)

1.22 MB JPG

>>3094996

This example uses a newer model, Anima Base V:1.0. The newer models are actually much better at not blending different characters together. Noticed I still kept the character descriptions as one sentence, and that all 3 boys are known characters that have a decent amount of training data:

Positive: masterpiece, best quality, score_9, score_8, score_7, newest, nsfw
male focus, 3boys, magazine cover, genshin impact, 16 year old teenagers, aether \(genshin impact\) with blond hair yellow eyes and single braid, scaramouche \(genshin impact\) with black short hair and purple eyes, albedo \(genshin impact\) with blue eyes and blond medium hair, bottomless, official costume, small penis, assisted exposure, spreading each others anus, erection, precum drip, testicles, male pubic hair, discarded clothing, seductive, seductive expression, take your pick, title "GENSHIN BUTTSLUTS", English text, barcode, <lora:anima_context_detailer_base10:0.5> <lora:AnimaMythD4rkL1nes:0.8>

Negative prompt: sweat, saliva, wet, worst quality, low quality, blurry, jpeg artifacts, watermark, text, error, deformed

The prompt is kept ambiguous with which character is actually performing which action, but the matrix is smart enough to bring it together.

File: 00096-369971951.jpg (1.87 MB, 2048x1536)

1.87 MB JPG

>>3094997

This example, made with the Anima checkpoint "Mindshift v: 1.0" was done as an experiment to see how far I could push the model to make different characters.

Positive: masterpiece, best quality, absurdres, high resolution, masterwork, detailed eyes, explicit, male focus, 1boy, solo
An illustration of a 16 year old teenage monk sitting with spread legs inside a Greek marble temple. The teenage boy monk is wearing leg and arm wraps and a white robe exposing his erection, testicles, male pubic hair, nipples, and pectorals. The 16 year old male monk has asymmetrical ginger colored and shaved hair and blue eyes. The monk is looking at the viewer with a serene expression. The teenage boy monk is surrounded by multiple Greek style white marble statues with bases. The statue on the left is a white marble statue of a teenage boy masturbating his erection. The middle statue shows a white marble teenage boy with curly hair kneeling with an open mouth and presenting his erection. The statue on the right is a marble adult fat satyr with beard and goat legs and goat horns with a erection and sagging testicles with beard. There are piles of gold coins and treasure chests around the temple. The temple is dark with dim lighting casting shadows. <lora:AnimaMythSmo0thL1nes:1>

Negative prompt: mature, lowres, (bad quality, worst quality:1.2), 6 fingers, 6 toes, bad eyes, bad pupils, bad iris, bad hands, bad fingers, long neck, bad anatomy, watermark, jpeg artifacts, ugly, poorly drawn

Notice that here, other than the first line, I used only complete sentences. I have found that anima works best when you skip most tags, and instead write a descriptive paragraph.

File: 00114-1477468871.jpg (1.3 MB, 1536x1536)

1.3 MB JPG

>>3094998

Try following the prompting structure of the examples to see how it can help. If there is a specific character or scenario you want to create, let me know and I can see if I can come up with a prompt for it.

One other thing I forgot to mention. There is an addon that will help you pick prompting words that the models were actually trained on, instead of guessing and hoping you are right:

>https://github.com/DominikDoom/a1111-sd-webui-tagcomplete

This is for Forge Neo user interface, which I use, but there are ones out there for Comfy as well:

>https://registry.comfy.org/nodes/ComfyUI-TagForge

These addons have a dataset of the tags used on Danbooru (and other booru) websites. The database for Danbooru was used to train illustrious (and pony), so those tags work best. As the words autocomplete, the number is the amount that tag has on Danbooru.

Good luck! and prompt for included picture, which used homosimileXL version 4:

Positive prompt: very awa, masterpiece, best quality, absurdres, newest, perfect quality, detailed background, intricate details, nsfw, realistic style, 2boys, from side, male focus, 19 years old, anal, rough sex, fucked silly, kaneki ken, tokyo ghoul, heterochromia, red eye, short hair, white hair, black sclera, black mask, slim body, pectorals, nipples, black pants, standing, small erection, testicles, body hair, worried, open mouth, heavy breathing, orgasm, cum, cum drip, motion lines, motion blur, looking at viewer, swirling faceless male shadow monster with black halo and demon wings, armor, pauldrons, greaves, groping,floating particles, light rays, abstract background, depth of field, <lora:cosmic_fantasy_art_illustriousXL:1>

Negative prompt: facial hair, injury, third eye, large penis, long penis, drool, saliva, breasts, pussy, twins, mirror_image, lowres, worst quality, bad quality, bad anatomy, signature, username, logo, watermark, artist name, old, bad hands

>>3094995

This one is so hot <3

File: 00138-4152411761.png (2.2 MB, 1728x1296)

2.2 MB PNG

>>3095010
Funny story, I did a whole series of like 60 pictures with a similar "caught" theme. CivitAI flagged them all and took them down. Even though I prompted for 19 years old and body hair, I think the school setting ticked them off. Oh well...

File: kira_8f7b16fa-1aa5-484a-a(...).jpg (324 KB, 1312x1760)

324 KB JPG

>>3094992
This is just a recommendation, but if you like inpainting and use comfyui I recommend krita, it works with anima (and practically any model), and it's amazing. Its really easy to config. I would put some examples but I'm not on my computer rn and from some reason 4chan don't let me upload pictures from my computer.

File: res.png (1.23 MB, 1360x768)

1.23 MB PNG

retard-kun here
still trying to get prompting right but results with loras is so much better. even anima lol

File: 2_HS2xDet_00006.jpg (1.99 MB, 1536x2720)

1.99 MB JPG

also me
just keeping thread alive

found this good captioner:
https://huggingface.co/Minthy/ToriiGate-0.5

combined it with https://huggingface.co/pixai-labs/pixai-tagger-v1.0 to make english language captions for all images in a folder automatically fast (feeds tags >.5 confidence to toriigate as a guideline)

it's good if you want to lazily train loras or if you wanna take a bunch of images that you like already and test out scenes like that in ai prompts

I vibecoded an auto setup script but it just gets flagged as a virus so I can't share it, but you can have chatgpt or claude make your own setup script if you paste the above probably. hope it helps someone

>>3093935
>>3094982
more Seifer please, he's fucking hot

File: 2.jpg (2.81 MB, 2720x1536)

2.81 MB JPG

>>3095071
still playing around with loras but it's still messy. anima is a good base for gens desu though

File: 3.jpg (2.9 MB, 2720x1536)

2.9 MB JPG

>>3095071
>>3095072

File: 4.jpg (2.5 MB, 2720x1536)

2.5 MB JPG

>>3095071
>>3095072
>>3095073

File: 00080-1361682995.jpg (1.18 MB, 1536x1536)

1.18 MB JPG

>>3095065

I posted 200 pictures of Seifer in 10 sets of 20 each here:

>https://civitai.red/user/gtgiygus/posts

You will have to go back about 40 days

This was the first set of 20:

>https://civitai.red/posts/30290126

All were done with that realistic style Lora that pretty much copied the game art from back in the 90s. Was very pleased with the look, but it was a bit stiff. Had really neat abstract backgrounds from the game art as well.

File: 00015-3080874532.png (1.98 MB, 1536x1536)

1.98 MB PNG

>>3095073
I like the emotion and energy here.

Keep working at it!

File: 3_HS2xDet_00008.jpg (2.22 MB, 2720x1536)

2.22 MB JPG

>>3095076
thanks

File: 3_HS2xDet_00016.jpg (2.05 MB, 2720x1536)

2.05 MB JPG

>>3095077

File: Messenger_creation_0AF80D(...).jpg (131 KB, 1216x832)

131 KB JPG

guess this is the only place where I can post homoslop
been thinking a lot about tsundere arjuna alter from fgo

File: 2_HS2xDet_00001.jpg (2.83 MB, 1536x2720)

2.83 MB JPG

>>3095076

File: 2_HS2xDet_00002.jpg (2.78 MB, 2720x1536)

2.78 MB JPG

>>3095115

File: 2_HS2xDet_00003.jpg (2.66 MB, 2720x1536)

2.66 MB JPG

>>3095115
>>3095116

File: 2_HS2xDet_00004.jpg (2.73 MB, 2720x1536)

2.73 MB JPG

>>3095115
>>3095116
>>3095117

File: image_00004.png (2.11 MB, 1080x1920)

2.11 MB PNG

Testing Krea again