VRAM 4GB で Flux.1 schnell
最小スペック云々見かけたのでさすがに動かないだろう・・・と思っていたら、意外や意外、動きました。
SDXLもそうでしたが、起動して最初にモデルを読み込む時だけ若干時間がかかりますが、あとは1024*1024で5～6分。
まぁ、オンラインサービスオンリーよりは選択肢が増えるという程度で。
※8/19追記
ちなみに512*512だと4分ちょい。サイズを小さくしてもそれほど速くはならない。
モデルをGGUF形式に変更し(モデルの読み込みは速くなります)、Q4で4分強、Q8で7分強でした。
ワークフローは下記より。
モデル名が変更されています。
ae.sft → ae.safesensors
flux1-schnell.sft → flux1-schnell.safesensors
https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main
作例
hamburger on dish,The word "Burger" is written in ketchup,
french fries, coca cola, tasty, food photography, dynamic shot
a beautiful cute joyful and playful 29 year old woman view from bottom,
red haired,dressed in a cozy sherpa jacket over a turtleneck and skinny jeans,
in the street of paris at night eiffel tower in background,
detailed masterpiece most beautiful artwork in the world Ultrarealistic,
Sony A7,Nostalgic lighting
A loving grandmother in her mid-70s reads a story to her grandchild in a cozy living room.
She has soft, silver curls, kind blue eyes behind half-moon glasses,
and a warm smile that crinkles the corners of her eyes.
She's wearing a comfortable cardigan in a soft pastel shade.
The child, whose face is not fully visible, sits on her lap,
pointing excitedly at the colorful picture book. Soft,
warm lamplight illuminates the scene, creating a heartwarming atmosphere.
photo of beautiful multiple Japanese girls, 20 year old age,
standing beach, happy face, black short haired, bikini swimwear,
detailed masterpiece most beautiful artwork in the world Ultrarealistic,
Sony A7, Studio lighting
3d model of a magical pokemon, game asset, detailed
beautiful illustration of a tattooed Geisha in a kimono,
blue short hair, detailed, beautiful red sun background
Fun, whimsical elements with bright colors and a sense of joy and creativity,
a 20-year-old woman, black short hair, wearing a red hoodie,
blue jeans and white sneakers, on the sign is written the phrase "Welcome",
in the style of cartoony characters, alone, figurative colorist, comiccore,
flat, nostalgic scenes, wide shot framing to capture the entire scene,
including the subject and the surrounding environment
Viral thumbnail of the black haired boy. in the style of cartoon realism,
digital art wonders, celebrity portraits, hikecore, appropriation artist,
strong facial expression, hedi xandt YouTuber.
Top yellow bold Text says “My Youtube Channel”.
He is sitting in front of her webcam. Behind him a YouTube plaque.
The room and everything else besides him is entirely realistic. only he is animated
Title: Display the title "Orina's Odyssey" in bold golden text, centered at the top of the poster, using a whimsical, serif font to evoke a sense of fantasy heroism. Add a subtle glow around the text to enhance its magical appearance while keeping it realistic. The elf's head should extend slightly above the title, creating a dynamic overlap.
Main Image: Present an ultra-realistic photograph of an female elf with red hair and golden eyes, standing in the center of the poster, framed in a medium shot. The elf is casting a bright light spell on murlocs. Her braided red hair is adorned with realistic-looking jewelry, and her golden eyes are illuminated by the spell's light. She is wearing detailed armor with a golden breastplate, weathered and engraved with intricate patterns. Her expression is determined and valiant, reflecting the tension of the situation. The scene is set in a magical forest, depicted in hyper-realistic detail, with towering trees, intertwined vines, and bioluminescent mushrooms emitting a soft, natural glow on the forest floor.
Background Elements: Integrate iconic elements from the Dungeons and Dragons universe, rendered in a realistic style, such as an ancient tower, partially hidden in the forest, visible through the trees, and mystical symbols carved into moss-covered stones. A dragon, barely visible, blends into the shadows of the forest, adding a subtle touch of menace. The color palette should remain dark and natural, with shades of emerald green and midnight blue to maintain a suspenseful atmosphere.
Secondary Characters: Add secondary characters from the World of Warcraft universe, rendered realistically, such as murlocs stealthily emerging from the river, their scales wet and glistening, with bright eyes in the dim light, and claws ready to attack. A spectral wolf, almost translucent, watches the scene from a distance, adding another layer of tension.
Visual Style: Ensure that the overall visual style is consistent with that of a realistic, epic movie poster. Use detailed photographic elements with natural yet dramatic lighting to capture the magical and heroic atmosphere of the story. The dark color palette should be illuminated by a golden light halo emanating from the elf's spell, accentuating realism while highlighting the heroic theme and dramatic intensity of the scene.
下記サイトの一番下のプロンプト入れてみたのですが、タイトル：～、メインイメージ：～、背景：～…といった指定も可能なんですね。
ただ、文字はちょくちょくスペルが被るのでタイトルなどはあとから合成した方が良い気もしますが。
レイヤー分割については下記サイトにも説明がありました。
プロンプトガイドを見たついでに・・・時間の経過の概念を試してみる。
A series of pictures depicting the evolution from monkeys to bipedal humans,
Wilderness background, colorful illustration
ちなみにこの画像はこちらのサイトで出力
上のプロンプトガイドにも無料で試せるGizAI Image Generatorへのリンクがありますが。
プロンプトが自然言語になっていくのが、Stable DIffusion ばかり使っていた利用者としては中々表現力のハードルが・・・。
この記事が気に入ったらサポートをしてみませんか？