趣味用

よろしく✌

[むぎたん @mugi_1633]

すごい！綺麗にモーションの収録出来てる！！

https://t.co/rU94whtWD2

[KaronPe🐻 @KaronPe_APL]

에이펙스 4:3 스트레치 해상도 쓸 때 좌우 감도는 바뀌는게 없는데 강제로 늘린 상태기 때문에 평균적으로 빠르게 느껴집니다 그럴 때 스팀 시작옵션에 +m_yaw 0.022 이것을 넣으면 16:9 비율의 해상도 감도랑 똑같은 느낌을 낼 수 있습니다

[Peng @pengsonal]

FOUND ANOTHER FREE LOCAL MODEL
UNCENSORED / ADULT-CAPABLE

FineP*rn V5 is a Krea 2 checkpoint for ComfyUI

• ~12GB INT8
• ~4.7GB text encoder
• no paywall
• realistic outputs

How to use:
1. download the model
2. add it to ComfyUI
3. load the workflow
4. generate

Model:
https://t.co/g4epy7IRgV

Fictional adults only. No real people or minors

[Peng @pengsonal]

FOUND AN UNCENSORED LOCAL VIDEO MODEL TO CREATE ADULT CONTENT 👀

NSFW Wan 1.3B was fine-tuned specifically for adult content

• text to video
• based on Wan 1.3B
• trained on NSFW data
• around 2.8GB for the newer checkpoint
• can run locally

don't download the old e10 checkpoint though

the repo now recommends: wan_1.3B_exp_e14.safetensors

check it out:
https://t.co/kahdzH9zBG

test it locally with fictional adult characters

[Rexo @rexsooo]

FOUND A LOCAL AI BODY SWAP MODEL

BFS can swap:

• body swap
• head swap
• face swap
• works with Krea 2
• Qwen Image 2.1
• Flux 2 Klein
• ComfyUI workflows included

the body swap version is already available

link: https://t.co/Wk7NPG4vBY

[Anogi_za @Anogi_otaAi]

誰にキスされたい？
俺はみんなにキスされたい😏

[さぶあいっちょ @sub_subai]

#乃木坂 #櫻坂 #日向坂 #青葉坂 #aiart

みーきゅん

未公開画像はnoteのメンバーシップで公���中だっちょ🍠こちらもよろしく〜🌸

https://t.co/aWjqTbQ57X

[Amelialtie @Amelialtie]

腰をヘコヘコさせながらだらしなく何度もイっちゃう眼鏡っ娘先生。「ダメ…イってるから…！んっ…！もうイってる！！」CV:陽向葵ゅか様（@yukanyan2525 ）の熱演をご堪能下さい💕

眼鏡っ娘先生のごほうびダイアリー２
DLsite https://t.co/hVOzHaAVbj
FANZA https://t.co/x0lKZ77ch5 #ad

[素人しかいない世界 @YuvrajS77173183]

マルチ参加するから●●させてくれ

[syato @syatoxgpi]

ぶいすぽっ　 一ノ瀬うるは　橘ひなの　千燈ゆうひ　花芽なずな　SFW #AIart #AIイラスト

[電ファミニコゲーマー @denfaminicogame]

「魔王城の隣」で村づくりするアクションRPG『ワロタ、俺の家のとなり魔王城www』早期アクセス版が11月13日に配信決定。体験版も近日配信予定
https://t.co/S1OqNA0W1B

素材を集めて家を建て、アイドル魔王のライブで熱狂した魔物たちの襲撃を迎え撃つ。TGS2026では実機試遊や試遊特典の配布も

[かんな@バーチャルJC @kanna3939]

MMD（MikuMikuDance）向けの、複数人数向けのモーションを探そうと思っても、専用タグが付いてなかったりでなかなか探しにくいので、調査した結果を順次紹介していきます。
ニコニコ動画で配布しているものです。

複数人数（２人以上）用モーション紹介 https://t.co/3vZUyzYyBJ

[任天堂株式会社 @Nintendo]

「Nintendo Switch 2 （日本語・国内専用） ゼルダの伝説 40周年 アニバーサリーエディション」は、10月29日（木）発売。
特別デザインの「Nintendo Switch 2 Proコントローラー」と「Nintendo Switch 2 キャリングケース」も同日発売。
#NintendoDirectJP #ゼルダの伝説40周年

[ドラゴンクエスト the DIVE【公式info】 @dq40th_dive]

【#DQ_DIVE 💧⚔】
⭐️大阪会場 開催詳細決定⭐️
 ドラゴンクエスト40周年記念展
 『ドラゴンクエスト the DIVE -まだ見ぬ冒険の舞台へ-』

🗓️開催期間：2027年1月26日(火)〜3月14日(日)
📍会場：グランフロント大阪 北館 ナレッジキャピタル イベントラボ

チケット先行販売が明日9月8日(火)10:00より、スクウェア・エニックス アプリでスタート🎊🎫
https://t.co/SBj0McHKVo
#DQ40th

[Killy @net_termina]

As promised, instead of for $15,000 GPUs, we max optimized a model that can fit comfortably in any GPU with 8GB VRAM running Windows

200 tok/s, *128k context, text/image/audio.

Anyone who wants to join the local AI community with 5 year old laptop can run this with recipe below, and this test was done on 3070ti laptop I forgot I had.

We pulled every instruct model released since May that fits 8 or 12 GB, and ran the six that mattered through the same 20 tasks: JSON extraction, code, a table, arithmetic, a fake-treaty honesty check, tool calls, language, a persona, a fact planted deep in a long prompt, and three photos. Gemma 4 E4B, Gemma 4 12B, Qwen3.5-9B, Qwen3.5-4B, Ornith-1.5-9B, the Qwen3.8-9B distill.

E4B is not the top scorer. Qwen3.5-9B and Ornith beat it on long context and on reading fine print in a photo. It won the pick for the things a newcomer on an old card actually needs: Apache 2.0, text + image + audio in, 3 GB of headroom on an 8 GB card, Weights Google trained to be 4-bit (QAT), and Google's own speculative drafter shipped next to them. Nothing else in the tier has all five, and the drafter alone is worth 2–3×.

[FHILY👑 @Oluwaphilemon1]

Qwen3.8-27B is running locally on an RTX 4060 with just 8GB of VRAM.

And it’s not some tiny quant running with a crippled context window.

The setup is:

• RTX 4060, 8GB VRAM
• Qwen3.8-27B
• Unsloth IQ4_XS quant
• 14.6GB model on disk
• 64K context
• Native MTP
• ~150 tok/s prefill
• ~5 tok/s decode
• Only 25 layers offloaded
• No VRAM overflow

The model is simply split between GPU and system memory to stay inside that 8GB VRAM limit.

And that’s the part I find fascinating.

A 27B model that you’d normally associate with much beefier hardware is now usable on a mainstream $300-class GPU.

It’s not fast by desktop-GPU standards. Around 5 tok/s for generation is obviously not going to win any speed contest.

But that’s not really the point.

The fact that you can load a model of this size, keep a 64K context window, use native speculative decoding, and actually run meaningful coding workloads on 8GB of VRAM is wild.

Then there’s the benchmark result.

Qwen3.8-27B reportedly scores 61.7 on SWE-Bench Pro, compared with 53.4 for Claude Opus 4.6 Max on the same benchmark.