AI art "Comm.DiTのテスト15" featuring #Kagamine Rin, #Vocaloid and #Megurine Luka
Generate New Artwork

Comm.DiTのテスト15

14の続きです。 13に追加して、説明書に記載のあった ※説明書とは、モデルおよびLoRA両方の説明書を読んでいます。主にLoRAの説明を反映させています。 1・LoRAの強さを0.9まで絵引き上げ。(前は0.7) 2・プロンプトに「anime coloring,」を追加。 13にくらべて、色彩とキャラがくっきりと表現されてきました。 設定を細かくいじるここで、ここまで変化するのか。と思いました。 Comm.DiTシリーズはユーザー様が調整したモデルとのことで、 調整したユーザー様の個性が色濃く出るのが面白いところです。 PixAIの運営さんが調整したTsubakiモデルとは、また別の遊び方だと 感じました。 PixAIさんのモデルは、細かい設定を省いてわかりやすく生成できる感じですが、 Comm.DiTは、細かくステータスがいじれる一方で、その分難度 が上がるモデルということなのかもしれません。 ただ、Sampling Stepsは低くても、きれいなイラストが仕上がるような作りもできるようなので、そこがモデルを作成される、ユーザー様の腕の見せ所なのかもしれません。 今回遊ばせていただいたモデルは、Sampling Steps12で、4枚生成で4200Kでした。コストが安価で、好きな感じのアニメイラストが作れました。 しっかり説明書があるモデルやLoRAだと分かりやすくていいなと思いました。 遊んでいて面白いです♪ Comm.DiT Test 15 Continuing from Test 14... Building on Test 13, I applied the following changes based on the documentation (I consulted the instructions for both the model and the LoRA, though I primarily incorporated the LoRA's guidelines): 1. Raised the LoRA strength to 0.9 (previously 0.7). 2. Added "anime coloring," to the prompt. Compared to Test 13, the colors and the character are rendered much more sharply. It really struck me how much of a difference fine-tuning the settings can make. Since the Comm.DiT series consists of models adjusted by users, it’s fascinating to see the creator's unique style shine through so clearly. It feels like a different kind of experience compared to the "Tsubaki" model adjusted by the PixAI team. PixAI’s own models feel designed for easy generation without needing complex settings, whereas Comm.DiT allows for granular control over parameters— which perhaps makes it a model with a steeper learning curve. However, it seems possible to produce beautiful illustrations even with low Sampling Steps, so that might be where the creator's skill really comes into play. For this test, I generated four images using 12 Sampling Steps at a cost of 4,200K. It was affordable, and I was able to create anime-style illustrations that really appealed to me. I appreciate models and LoRAs that come with clear documentation; it makes the process much easier to understand. It’s really fun to play around with! ♪

#Kagamine Rin#Vocaloid#Megurine Luka
More to See

Commenti

2girls, kagamine rin\(vocaloid\), megurine luka\(vocaloid\), sitting, cafe, talking, looking at each other, happy, strawberry shortcake, cheesecake, anime_screenshot, anime coloring, kagamine rin: solo, blonde hair, short hair, blue eyes, petite, hair ornament, hairclip, detached sleeves, black shorts, sailor collar, skin indentation, bow, shirt, leg warmers, megurine luka: school uniform, hair ornament, hairclip, brooch, black dress, sleeveless dress, wrist cuffs, black thighhighs, asymmetrical footwear, thigh boots, black footwear
Mostra parametri
Immissione richiesta originale
Character A and Character B are sitting across from each other in a café. They are chatting happily while looking at each other's faces. There is a strawberry shortcake in front of Character A. There is a cheesecake in front of Character B. Character A:,kagamine rin, solo, blonde hair, short hair, blue eyes, Petite, hair ornament, hairclip, detached sleeves, black shorts, sailor collar, skindentation, bow, shirt, leg warmers, Character B:,Megurine Luka Sour uniform,hair ornament,hairclip,brooch,black dress,sleeveless dress,wrist cuffs,black thighhighs,asymmetrical footwear,thigh boots,black footwear, anime_screenshot,anime coloring,
Dimensione
1536 x 1536
Passaggi di campionamento
12
Metodo di campionamento
Euler
Scala CFG
1.1
Negativo
worst quality, photorealistic, bad anatomy, blur, low resolution

Modello e LoRA utilizzati