為什麼做這次評比?
Krea2 開源之後,社群微調版本開始冒出來。我這次拿了四個微調模型,跟官方 Turbo 打一場公平的擂台:
- 官方 Turbo(
krea2_turbo_fp8_scaled)— 基準 - Kreamania variant3 / variant4 / variant5 — 同系列三個版本(INT8 量化)
- Muse by Stable v2.0 Turbo(INT8)— 據說擅長時尚街拍與商業廣告
公平比法:35 組 prompt × 每組固定 seed,五個模型看到的文字與 seed 完全相同,只換 diffusion model。Prompt 覆蓋插畫、水墨、油畫、黑白攝影、商品攝影、時尚街拍、動畫等 20 種以上風格域,一個晚上跑完 175 張。
快速結論
- variant3 最有藝術感,出圖有 Midjourney 的味道——它會「詮釋」你的 prompt 而不是照抄,黑白攝影裡冒出紅色草莓這種「將錯就錯」反而成了整批圖的驚喜
- Muse 的商業攝影傳言屬實:時尚街拍、香水、腕錶、美妝五組商業題全部由它拿下或並列第一,皮膚質感最自然
- 官方 Turbo 是風格服從性之王:低多邊形、字符肖像、2D 扁平動畫這些「圖形類」題目,微調版全都會偷渡寫實感,只有官方版忠實還原
- variant5 細節密度最高、帶動漫感,商品攝影的水花戲劇性最好,但要留意細節幻覺(它幫球鞋長出了 Nike 勾勾)
- variant4 最穩定中庸,錯誤率低,但構圖偏靜態
- 四個微調全是 int8_tensorwise 量化——ComfyUI 要升到 0.28 以上才跑得動,舊版會直接報
KeyError: 'int8_tensorwise'
測試方法
- ComfyUI v0.28.3、RTX 5090、1024×1024、8 steps / CFG 1 / euler / simple
- 管線帶 Qwen3-VL 提示詞擴寫(固定 seed,五模型拿到的擴寫文本逐字相同)
- 對照圖從左到右固定順序:turbo → variant3 → variant4 → variant5 → muse
variant3:黑白照片裡的紅草莓
先看讓我改觀的一組。Prompt 明確要求「black and white photograph」,一位女子捧著木碗草莓。

Prompt:
A black and white photograph of a woman with light hair loosely braided, leaning forward to press her cheek against a wide wooden bowl heaped with ripe strawberries, her face partially hidden by the fruit, one eye visible and gazing directly at the viewer, wearing a thin hoop earring and a fine chain bracelet, hands cradling the bowl from below, set against a softly blurred orchard background, natural daylight illuminating the scene with high contrast and grainy texture, shallow depth of field emphasizing the subject and the mound of berries.
四個模型乖乖交出黑白照。只有 variant3 把草莓染回鮮紅——嚴格說是「沒聽話」,但這個選擇性上色讓整張圖從習作變成了作品。這正是 Midjourney 式的魅力:模型有自己的美學主張。
同樣的「詮釋力」也出現在蘭花遮眼的超現實人像:

Prompt:
A hyper-realistic, high-contrast black-and-white fine art portrait capturing a surreal fusion of human and nature. The composition is an intimate close-up of the lower face and neck of a woman, her eyes completely hidden behind a massive blooming orchid. Her skin is smooth with visible texture - pores, faint freckles, the soft line of her jaw. Her lips are relaxed and slightly parted. Covering the upper half of her face is a giant moth orchid, its broad translucent petals revealing delicate vein structures and speckled markings like scattered ink. The flower's intricate central column sits over the bridge of her nose, a striking focal point, and the petals drape down her cheeks as if growing from the skin itself. The mood is ethereal, dreamlike, and faintly melancholic - a visual poem about beauty, transformation, and the blurred boundary between the human form and the natural world. Dramatic chiaroscuro lighting from the left models the petals and lips while the right side falls into deep shadow; pure black background isolates the subject. Shot on medium format film with rich grain and tonal depth, vertical portrait framing, microscopic detail in the flower's veins and the skin's texture.
variant3(左二)給了唯一的側臉構圖,唇色微微滲出暖調,花瓣的光影最像底片。其他四個都是標準正面構圖——正確,但少了故事感。
第三組,漂浮在雲海上的青銅面孔。variant3 直接把「漂浮面孔」重新詮釋成帶肩頸的雕像半身——偏離最遠,卻也最有畫面張力:

Prompt:
An ethereal realm where faces cast from ancient bronze, adorned with coral filigree and luminous jade patterns, float above an ocean of slow-rolling clouds. Their serene expressions reflect forgotten dreams, while ribbons of golden kelp entwine their forms, pulsing with warm light. Comet trails streak the deep indigo sky, casting otherworldly hues upon giant floating monoliths that hum silent chords. Every detail is rendered in ultra-HD photographic clarity, capturing the verdigris texture of the bronze and the mystical radiance of the jade inlay, a surreal scene where time and space dissolve into a symphony of light and fantasy. Moths with crystal wings drift past, leaving trails of glowing dust, adding fleeting magic to the dreamlike composition. The contrast between burnished gold and deep indigo creates a hypnotic visual tension, inviting the viewer into a waking dream of supernatural beauty.
氛圍光影也是它的主場。雨夜街燈這組,variant3 的燈暈和濕石板反光是五張裡唯一有「電影劇照」感的:

Prompt:
A solitary figure walks barefoot down a rain-slicked cobblestone street at dawn, wearing a long charcoal coat and a deep hood that hides their face, with wind-blown pale hair and wet maple leaves clinging to the hem; the scene is lit by a row of dim iron street lamps glowing against a cold violet-grey sky, dripping bare chestnut trees lining the road behind, captured in cinematic realism with shallow depth of field and atmospheric haze.
variant3 小結:藝術感最強、光影最會演,但代價是提示詞紀律最差——嚴格的風格約束(純黑白、特定媒材)它會自己改戲。接案交件慎用,自由創作是寶。
官方 Turbo:被低估的風格服從性
微調模型有個共同副作用:往寫實攝影的方向漂移。低多邊形狐狸這組看得最清楚。

Prompt:
A poised fox stands alert in the foreground, rendered in geometric polygonal facets of deep crimson, rust, and sharp accents of ice blue and cream, its tail sweeping low as it faces right; behind it, a massive pale silver moon dominates the sky, composed of layered textured panels that radiate cool light against a backdrop of stylized pine silhouettes in charcoal and deep teal; tall frost-tipped grasses with steel-blue blades rise from the foreground, their forms built from angular, brushstroke-like segments, all under a dramatic gradient sky shifting from midnight blue to pale silver, evoking a still winter night; the entire scene is executed in a mosaic-like digital painting technique with visible texture strokes mimicking hand-cut paper or fabric, creating depth and tactile richness without adding new elements or characters.
官方 Turbo(最左)的切面最乾脆,最接近「幾何多邊形」的字面要求;variant4 次之。variant3 直接畫了一隻寫實毛髮的狐狸站在多邊形背景前——美,但已經不是這道題了。variant5 和 Muse 介於中間,毛髮質感悄悄長回來。
字符堆砌人臉、2D 扁平動畫這兩組也是同樣的劇本:官方 Turbo 忠實,微調版「偷渡寫實」。如果你的工作流常跑圖形化、插畫類風格,官方 Turbo 還是最可靠的選擇。
全員高水準的一組:黑暗奇幻油畫
值得欣慰的是,Krea2 的底子讓所有微調在「油畫敘事」這種主流題材上都很能打。沼澤先知與貓頭鷹議會這組,五張都可以直接當書封:

Prompt:
Category: Dark Fantasy Oil Painting / The Seer of the Drowned Marsh and Silent Rite. Era/Style: Romantic Era Gothic Art, reminiscent of sublime landscape painters mixed with the darkest late-period murals, heavy impasto for reeds and feathers, glazing for atmospheric mist, dramatic chiaroscuro, authentic aged canvas texture. Detail level: extreme macro detail on the slick texture of drowned willow roots, the layered softness of owl plumage, the viscous ripple of black marsh water around a stone basin, and the wild braided hair of the seer. Sharp focus on the offering basin with deep background fog blur. Composition: wide landscape orientation; the seer stands at the base of a half-sunken, hollowed willow, a half-circle of great horned owls perched on the roots around her; dense fog banks and a bone-white moon fill the background; low horizon line emphasizes the leaning trees and the smallness of the figure. Subjects: an aged seer with skin like river clay and eyes that glimmer faintly silver, robed in stitched reed-cloth and oiled hide, hair braided with fishbones and dried sedge, tipping a shallow stone basin of dark water with both hands, her expression fierce with concentration; five large owls watch the rite, feathers unusually glossy under the pale moonlight, one holding a small tarnished key in its beak. Color palette: dominant deep umber and moss green, charcoal black shadows, bone white moon, cold silver accents; high contrast between the dark marsh and the glowing sky. Lighting: primary light from the pale moon casting a cold monochrome glow through the fog; a faint secondary glimmer from the seer's eyes; deep shadows swallow the waterline; chiaroscuro emphasizes the mystery. Texture: rough cracked bark, downy feathers, viscous water, soft blended fog edges, crisp detail on the basin and the seer's face, visible brushstrokes in the background, aged canvas grain. Hyper-realistic dark fantasy oil painting, primal quiet-horror aesthetic, sense of ancient watchfulness, professional classical painting quality.
差異在細節個性:variant4 構圖最近最滿、variant5 的服飾紋理堆得最密、Muse 的月光最乾淨。這種題目怎麼選都不會錯。
順帶一提 variant5 的「動漫感」——水晶蓮花這組 2D 動畫題,它的線條和上色最接近日系插畫的完成度,喜歡動漫風的人可以特別留意它:

Prompt:
Radial 2D Japanese anime space panorama with hairline draftsmanship and coral-violet color wedges: an adult anime heroine depicted as a moon-haired guardian stands at the heart of a kilometer-wide crystal lotus. The camera looks obliquely down its geometric crown, where glass petals sweep past the viewer and a convoy of tender ships dwindles toward the vanishing point. Four pale pinions brace her against stellar wind as both palms weave charged pollen into a missing navigational chord. A dusk-violet pressure suit makes her small figure readable among countless transparent petal ribs. Coral lamps burn inside loose petals; cold violet illumination outlines her hair and shoulders. By conducting the particles like an orchestra, she opens one temporary passage through a blank sector before the lotus disperses. Rendered as a meticulous long-form anime illustration built from flat chromatic planes.
Muse:商業攝影棚體質
傳言說 Muse 擅長時尚街拍與商業廣告。五組商業題跑完,傳言屬實。
時尚街拍

Prompt:
Candid off-duty fashion street snap of a tall woman striding across a zebra crossing in a European city at midday, caught mid-step with one hand holding an espresso cup and the other tucked in the pocket of an oversized camel wool coat worn open over a graphic white tee, wide-leg pinstripe trousers and chunky white leather sneakers, slim rectangular sunglasses, hair in a loose low bun, a structured black shoulder bag swinging with her stride; shot like a paparazzi-style 35mm frame with slight motion in the coat hem, blurred taxis and blurred pedestrians behind her, natural overcast light, true-to-life color, fine film grain, sharp focus on the subject, authentic streetwear editorial energy.
Muse(最右)的步態抓拍、大衣擺動、街景層次最像真的雜誌街拍頁。variant5 這組把直紋西裝褲換成了米色燈芯絨——時尚題的服裝忠實度它最不穩。
香水廣告

Prompt:
A luxury commercial product photograph for a perfume campaign: a faceted glass bottle filled with pale amber liquid stands on a wet black marble slab, a crown of fine water droplets frozen mid-splash arcing around it, a single sprig of orange blossom resting at its base; deep charcoal studio background with a soft golden gradient glow behind the bottle, crisp twin rim lights tracing the glass edges, razor-sharp focus on the bottle with the splash caught at high shutter speed, controlled reflections on the marble, advertising-grade cleanliness and polish, hyper-detailed liquid and glass rendering, elegant negative space at the top of the frame.
variant5 的皇冠水花最戲劇化,Muse 的整體打光最接近可以直接交付的商業稿,官方 Turbo 反而最保守(水花最小、瓶身最小)。
美妝封面——順便揭露一個問題

Prompt:
High-fashion beauty editorial close-up of a model with luminous deep-brown skin, slicked-back hair, and sculpted cheekbones, eyes closed mid-blink, wearing glossy tangerine lips and a single line of silver foil across one eyelid; lit by a split of two colored gels, soft magenta from the left and cool cyan from the right, meeting in a clean seam down the center of the face; bare shoulders against a seamless plum studio backdrop, dewy skin texture with visible pores and fine highlights, ultra-sharp macro detail on the lips and foil, glossy magazine-cover finish, bold minimal composition with generous headroom.
這組 prompt 指定了 deep-brown skin(深棕色皮膚)。結果只有官方 Turbo 忠實呈現,四個微調版全部把膚色調淺了。微調資料集的偏差在這裡現形——如果你的創作需要準確的深膚色人像,微調版要多加小心。
另一個商用警訊:運動廣告這組,variant5(右二)自己幫球鞋畫上了 Nike 勾勾——prompt 裡完全沒有提到任何品牌。商業出圖記得檢查有沒有 logo 亂入。

Prompt:
Dynamic commercial sportswear campaign shot of a sprinter exploding off the blocks on a rooftop running track at dusk, frozen at full extension with chalk dust bursting from his fingertips, wearing a matte storm-grey compression suit with volt-green seams and matching spike shoes; low camera angle from track level, dramatic warm rim light from the setting sun carving his silhouette against a deep blue city skyline, a second cool fill light from the front, sweat and chalk particles glittering in the backlight, crisp motion tension in every muscle, cinematic advertising color grade with strong contrast, wide negative space on the right of the frame for campaign copy.
結論:哪個場景用哪個模型?
| 需求 | 首選 | 備選 |
|---|---|---|
| 藝術感、驚喜、Midjourney 味 | variant3 | variant5 |
| 時尚街拍、商業廣告、雜誌質感 | Muse | variant5 |
| 寫實人像皮膚質感 | Muse | variant4 |
| 插畫、圖形化、2D 動畫 | 官方 Turbo | variant4 |
| 動漫感插畫 | variant5 | variant4 |
| 嚴格遵守 prompt(交件用) | 官方 Turbo / variant4 | — |
| 奇幻細節、商品攝影 | variant5 | Muse |
我自己的用法:日常預設官方 Turbo,想要「給我驚喜」的時候切 variant3 多抽幾張,商業人像案切 Muse。五個模型共用同一個工作流,只要換 UNETLoader 裡的模型檔,切換零成本。
其餘 prompt 收錄在文末附錄,歡迎自行測試。這批測試集我也會繼續用在 Boogu、Ideogram 4、Z-Image、Qwen Image 2512 的評比上——敬請期待。
附錄:其餘 24 組 Prompt 分享
正文出現過的 11 組不再重複,以下是其餘 24 組,附五模型對照圖(左到右:turbo → variant3 → variant4 → variant5 → muse),歡迎自行測試。統一設定 1024×1024、8 steps、CFG 1(Krea2 Turbo 系參數),括號內為本次使用的 seed。我的工作流有掛 Qwen3-VL 提示詞擴寫,直接用原文生成的構圖會略有不同,但風格結論不變。
01. 插畫/繪本(seed 41207)

A lively comedic scene on a weathered wooden jetty with a red-and-white striped lighthouse in the background, soft early-morning light over a glassy green lake. Three raccoons caught mid-chaos in one frame: (1) a proud raccoon standing upright, gripping a bowed fishing rod with both paws, hauling a huge shimmering trout out of the water, eyes gleaming with triumph. (2) behind him, a sneaky younger raccoon creeping away on tiptoe with a stolen sandwich clamped in its jaws, having just raided an open picnic basket, glancing back with a guilty smirk. (3) at the edge, a plump old raccoon tangled head-to-toe in fishing line, one hind leg hoisted by the taut string, mouth open in an exaggerated squawk of outrage. Dynamic slapstick composition, expressive cartoon-like faces, motion and splashing water, fur flying, warm cinematic lighting, highly detailed, playful storybook illustration style, vivid colors, sharp focus.
02. 水墨奇幻插畫(seed 52318)

A lone young wanderer with short copper hair, wearing a patched traveling cloak and worn leather boots, kneels on cracked desert clay, gazing upward at the colossal skeleton of a whale-like leviathan arched overhead, its ribs bleached and banded with corroded iron rings; a small desert fox sits calmly beside the wanderer's pack, and a cracked ceramic water jug lies tipped over near their knees, rendered in a detailed, muted palette dominated by sepia, bone white, and dusty ochre under soft, diffused daylight, with subtle ink-style linework and shading evoking a classic illustrated fantasy tone.
03. 微距商品攝影(seed 63429)

An extreme macro studio photograph centered on a glossy black lacquer theater mask fractured open down the middle. Emerging from the pale fissure inside the mask are dozens of tiny, hyper-detailed silver clockwork gears interlocked with blooming white lilies. Liquid mercury drips from the lily petals into the crack. Set against a deep charcoal background. Harsh directional rim lighting highlights the sharp cracked edges of the lacquer and the cold sheen of the silver gears, creating extreme visual contrast and deep shadows.
04. 維多利亞風俗畫(seed 74530)

A masterfully rendered artistic illustration in the classic 1892 Victorian social realism style with influences from period photography, characterized by detailed figure rendering, textural shading, and elegant posing typical of late 19th century genre art. The composition centers on four women playing croquet on a garden lawn beside a conservatory, using group dynamic and period clothing to convey social life. The women wear high-collared summer dresses and straw hats, mid-swing and conversing. The artistic technique employs soft tonal shading for fabric textures, ribbon details, and facial features, precise linework for mallets and wickets. Every element faithfully preserves the 1892 aesthetic with exact reconstruction of the poses, clothing, garden furniture, and outdoor setting, evoking the leisurely customs of the Victorian era with masterful illustrative techniques of the period.
05. 扁平編輯海報(seed 85641)

A bold editorial art poster of a young man in an emerald green trench coat, surrounded by oversized golden sunflowers and torn geometric paper shapes, layered screen-print textures, playful asymmetrical composition, vivid teal, coral and lemon-yellow palette, gallery-quality contemporary illustration, subtle paper grain.
06. Pulp 封面插畫(seed 96752)

Pulp noir illustration, retro sci-fi, desaturated tones, heavy shadowing, bold ink lines, cinematic lighting, frontier grit, vintage magazine cover. A towering, heavily muscled wolf figure - a barbarian titan in form - stands center stage, his grey fur matted with ash and old scars, coarse as frayed rope. He wears segmented iron-plate armor edged with dull copper rivets, and a rust-red mantle drapes over his shoulders, absorbing the dim light like shadow-stained cloth. His muzzle is set hard, eyes narrowed in grim resolve. Beside him, a young male scout with wind-burned skin and cropped dark hair kneels on cracked, barren ground, his gear utilitarian: earth-toned patched fabric, reinforced joints, straps worn thin. He grips a short hooked blade, its handle carved with angular clan markings, gleaming faintly under the pale sky. In the foreground, a massive dented war-helm lies half-buried in grit, its crest etched with an unknown sigil, surrounded by snapped spears and corroded blades. The setting is a desolate alien frontier, dust drifting in slow motion under a sepia sky, low sun casting long, jagged shadows. Rendered with bold ink lines, heavy chiaroscuro, and gritty texture, evoking early 1940s pulp sci-fi covers. Muted browns, greys, ochres, and faded greens dominate, no bright accents. The composition centers the wolf and scout against a vast empty wasteland. Mood is tense, foreboding, cinematic, and mythic.
08. 情緒人像攝影(seed 21897)

A man with tousled dark hair sits on a worn leather armchair against a bare plaster wall, cradling a chipped enamel coffee mug in both hands with steam curling upward; he wears a faded olive t-shirt and dark trousers, looking directly at the camera with a calm, contemplative expression, illuminated by soft natural light filtering through horizontal blinds on a window to his left, casting directional shadows across his face and body, rendered in a cinematic, moody photographic style with high contrast and grainy texture.
09. 黑白攝影(seed 32908)

A black and white photograph of a tiny fluffy kitten sleeping peacefully with its head resting on the round keys of an old manual typewriter, its paws gently draped over the key rows, while a half-typed manuscript page curls up from the platen above; soft natural light streams from the right, casting gentle shadows across the machine's worn enamel body and highlighting the kitten's fur and the texture of the keys, creating a quiet, nostalgic, and tender atmosphere.
13. 數位插畫/概念(seed 76342)

A massive bronze raven head with intricate riveted and weather-cracked plating hovers above a desolate, misty salt flat under an ashen sky, its hollow eye sockets glowing with cold blue fire and a thin beam of light spilling from its open beak down to a solitary human figure standing in the shallow mirror of water below, wearing pale robes and facing the raven, with distant flat-topped mesas and sparse dead brush along the horizon, rendered in a cinematic digital illustration with dramatic chiaroscuro lighting and a cold, oppressive atmosphere.
14. 表現主義混媒(seed 87453)

An eerie, high-contrast black-and-white portrait of an elderly Sicilian fisherman standing waist-up on a wind-scoured shore, clutching a bundle of frayed netting, face carved with deep wrinkles and eyes closed, weathered neck taut with quiet dignity; dramatic Rembrandt-style chiaroscuro sculpts his gaunt features and hollow cheeks, heavy textured impasto and visible expressionist brushstrokes overlay the image like cracked paint on weathered canvas, subtle mixed media of sand, salt crust and scratch marks, storm-swept sky and blurred horizon behind him, gnarled hands clasping the net cords, worn dark oilskin coat with a simple collar, melancholic, timeless and cinematic mood, tactile oil-paint texture, high detail on skin pores and veins, soft vignette, muted tonal range, surreal gothic atmosphere, expressionist composition, evoking memory, solitude, and quiet resilience.
15. 表現主義人像(seed 98564)

A painterly, expressive scene of a poised woman with a sleek copper-red bob, three-quarter view, head-and-shoulders composition, wearing a structured navy blazer over a cream blouse and a single thin gold hoop earring, soft coral lips and subtle bronze eye makeup, natural skin texture with fine pores and a faint blush, thoughtful and resolute expression with soft catchlights in her eyes, a trace of melancholy in the gaze. Bold, loose brushstrokes, thick impasto and visible canvas texture, layered acrylic and gouache mixed media with ink scribble marks and collage fragments. High saturation palette: teal shutters, warm apricot background, deep violet shadows below. Composition balanced vertically, a window ledge in the foreground, distant chimneys and wires suggested with rapid gestural marks, abstracted urban details. Emphasize contrast of flat color blocks and energetic brushwork, soft daylight, contemplative mood, textured edges, slightly simplified anatomy, modern expressionist aesthetic, cinematic perspective, studio painting quality.
17. Kodachrome 復古(seed 20786)

A nostalgic vintage kodachrome slide photograph from the 1960s, shot on classic Kodachrome color film, featuring rich saturated colors with the signature warm color shift, distinctive visible film grain and texture, natural daylight with soft shadows, sharp focus, and an authentic mid-20th-century retro aesthetic. A stylized, elegant woman with a short glossy black bob is depicted in side profile facing right, eyes lowered and expression serene. Her skin is pale, contrasting with her dark hair and the vivid colors around her. She wears a fitted emerald-green cheongsam with a high collar and fine silver embroidery, and in her right hand she holds an open oiled-paper parasol painted with plum blossoms, her left hand resting lightly at her side. The background is a rich, textured wash of jade and gold hues with subtle cloud patterns that add depth and warmth to the composition.
18. 裝飾人像(seed 31897)

A highly detailed, intricate profile portrait of a woman in side view, her face and neck adorned with elaborate copper wire lattices that flow like woven lace and curling tendrils, partially veiling her features while revealing her eyes, lips, and nose; the background is dark and textured with teal smoke drifting upward behind her head, contrasting against a muted slate backdrop; the lighting is soft and directional, catching the burnished sheen of the copper work and casting subtle shadows that emphasize depth and texture; the composition is tightly framed on the subject's head and shoulders, grounding the image in quiet mystery and artistic elegance.
19. 印象派海景(seed 42908)

A misty, atmospheric harbor scene at first light, featuring an old wooden fishing trawler moored in still water, its hull painted in flaking greens and rust reds with a warm amber lantern glowing from the wheelhouse, casting a soft reflection on the fog-dappled surface; the trawler's mast and boom are partially swallowed by dense fog that erases the background, leaving only faint silhouettes of pilings and a distant crane; a solitary figure leans on the rail near the bow, barely visible against the gloom; the foreground includes overturned dinghies resting on the shingle; the composition is rendered in impressionistic brushstrokes with soft, diffused lighting, evoking a quiet, melancholic mood; the palette is dominated by muted greys, sea greens, and deep umbers, with the lantern glow providing a focal point of warmth.
20. 寫實人像攝影(seed 53019)

A close-up portrait of a young man with loose dark curls, a faint scatter of freckles across his nose and cheekbones, and striking hazel eyes looking directly at the camera with a neutral expression; he wears a soft, textured sage-green linen shirt with an open collar, set against a plain, softly lit background with gentle shadows cast by natural light from one side, emphasizing the contours of his face and the weave of the fabric, in a realistic photographic style with high detail and shallow depth of field, capturing an intimate, contemplative mood.
21. 金色時刻街景(seed 64120)

A sun-washed whitewashed island lane framed by a low stone archway, cascading jasmine and scarlet geraniums spilling from window boxes overhead, leading to a weathered cobalt-blue double door with hand-forged iron hinges beneath a rounded transom window, flanked by a faded turquoise shutter on the left and a narrow gate on the right, terracotta pots resting on the steps, scattered petals on the smooth flagstones, warm golden-hour lighting casting long shadows, realistic textures of lime-plastered walls, worn stone treads, and climbing vines, photorealistic detail with soft depth of field emphasizing the blue doorway as the focal point.
22. 親密人像攝影(seed 75231)

A close-up portrait of a young Asian woman with chin-length, softly waved black hair tucked behind one ear, looking directly at the camera with gentle eyes and a subtle, calm smile; she wears an oatmeal-colored knit cardigan over a white camisole, a hint of natural freckles on her cheeks, posed with her head tilted slightly toward her left shoulder, in a softly lit, neutral-toned room that keeps attention on her features, captured in a high-resolution photograph with shallow depth of field, warm ambient window light, and natural skin tones, conveying an intimate, serene, and elegant mood.
23. 蒸汽龐克 sepia(seed 86342)

A man in profile wearing a detailed steampunk aviator helmet fitted with brass goggles, pressure valves, and a small altimeter dial, standing against an industrial backdrop of mooring towers and drifting airships trailing white steam; he wears a high-collared canvas flight jacket with ornate stitched insignia on the sleeve and linked brass chains across the chest, rendered in sepia tones with dramatic chiaroscuro lighting, a photograph of a solitary figure in a soot-hazed airfield landscape, medium shot, slightly off-center composition, soft focus background emphasizing the subject's mechanical headgear and the long dark scarf thrown over his shoulder, no additional characters, grounded in realism with artistic stylization.
25. 四格複合構圖(seed 18564)

A four-panel composite featuring a woman in profile wearing a wide-brimmed hat, presented in three distinct artistic interpretations alongside her original black-and-white photograph. The top-left panel is a realistic photographic portrait of the woman in a dark tailored jacket with a silk scarf, captured in side view against a plain background. The top-right panel is a stylized geometric illustration using bold, flat colors - orange, teal, cream, and black - with sharp angular forms and visible brushstroke textures, keeping the same profile and attire. The bottom-left panel is an abstract expressionist interpretation dominated by large teal, cream, and black shapes with dynamic lines and minimal facial features suggesting the profile, the scarf rendered as a sweeping orange ribbon. The bottom-right panel is a vibrant layered abstract composition of intersecting planes, circles, and radial lines in plum, orange, teal, and gold, with a central circular motif of concentric rings and dots, set against a white background with the same profile silhouette implied through fragmented geometry. All panels share the same subject, consistent profile framing, and attire, differing only in style from photorealism to abstraction, no additional objects or text.
26. 超寫實動態攝影(seed 29675)

A tiny copper clockwork robot with exposed ticking gears and jointed spindly limbs, perched precariously on the back of a real greyhound mid-stride down a dusty dirt racetrack, both moving dynamically as if in motion, faint wisps of steam venting from the robot's shoulder joints, the dog's brindle coat streaked with dust and sweat, the track littered with paw prints and loose gravel under harsh midday sun, shallow depth of field blurring the background into a hazy expanse of earth and distant railing, the robot's single lens eye glowing with focused amber light, the greyhound's ears pinned back and tail rigid with effort, no other characters or objects present, the scene rendered in hyper-detailed photorealistic style with cinematic lighting that emphasizes texture, motion blur, and gritty realism.
27. 字符肖像/賽博(seed 30786)

A striking digital art piece in a dark cyberpunk and glitch-art style. The image depicts a human face emerging from pure darkness, constructed entirely from cascading lines of pale monospaced computer code, the characters forming the contours of the nose, cheeks, and forehead, creating a ghostly, fragmented appearance that dissolves into the black background at the edges. The focal point: the eyes are piercing, luminous amber with intricate iris detail and sharp reflections, staring directly at the viewer - organic and alive in haunting contrast with the digital text around them. Typography-based portraiture: the face is built from characters, numbers, and symbols rather than drawn lines. High-contrast chiaroscuro lighting, the only light sources the glowing amber eyes and the faint luminescence of the text itself against a pitch-black background. Atmosphere mysterious, technological, and slightly unsettling, evoking a digital consciousness waking up. Small floating particles of digital dust surround the head, and the text density varies to sculpt shadows and highlights across the face.
28. 畫意超寫實動態(seed 41898)

A flamenco dancer mid-turn on the parapet of an old stone rooftop against a dramatic, storm-darkened sky with churning slate clouds; she wears a tiered crimson dress with black lace at the bodice and fitted sleeves, paired with low-heeled dance shoes; her dark hair whips around her as one arm arcs overhead and the other sweeps the skirt wide, chin lifted in fierce concentration, one heel striking the weathered stone ledge; cinematic lighting rims her silhouette against the moody backdrop, emphasizing motion and defiant grace, rendered in a painterly, hyperrealistic style with soft shadows and rich textures, capturing a moment of suspended intensity.
30. 奇想木雕特寫(seed 63010)

An ornate wooden clock reimagined as an owl clock hangs on the warm stone wall of a cozy cottage, bathed in soft afternoon golden light that spills across the grainy wood texture. At the precise moment the bells chime, downy owlets emerge from an arched window above the dial and hop merrily along a small carved balcony. One owlet flutters mid-air with wings spread wide while another bobs gently, tufted ears tilted like quarter notes drifting in the air. The casing replaces traditional leaf carvings with acorn and moth wood sculptures, and the weights are tiny brass lanterns dangling from copper chains. Golden hour light bleeds through the open window, casting long shadows against the facade in a fantasy book-like atmosphere. The lens focuses on the detailed downy texture of the owlets with shallow depth of field, capturing their curious poses as if caught in time. Exquisite craftsmanship marks every carved detail, from the intricate wood grain to the polished little lanterns.
35. 腕錶生活風格廣告(seed 28565)

A premium lifestyle advertising photograph of a men's dress watch with a brushed steel case, midnight-blue sunburst dial and dark brown alligator strap, worn on the wrist of a man adjusting the cuff of a charcoal flannel suit jacket over a crisp white shirt; tight macro crop on the wrist and hands against the blurred warm bokeh of a hotel bar in the background, dramatic window light raking across the dial to reveal the sunburst texture and polished indices, shallow depth of field, immaculate grooming and fabric detail, sophisticated editorial-commercial mood, rich contrast with deep shadows and controlled highlights.
模型下載
官方 Krea2 Turbo(基準)
Hugging Face:https://huggingface.co/Comfy-Org/Krea-2
| 檔案 | 放置位置 |
|---|---|
krea2_turbo_fp8_scaled.safetensors | models/diffusion_models/ |
qwen3vl_4b_fp8_scaled.safetensors | models/text_encoders/ |
qwen_image_vae.safetensors | models/vae/ |
Kreamania(variant3 / variant4 / variant5)
Civitai:https://civitai.com/models/2729631/kreamania
Muse by Stable Yogi v2.0 Turbo INT8
Civitai:https://civitai.com/models/2741166/muse-by-stable-yogi-krea2?modelVersionId=3131430
重要:Kreamania 三個版本與 Muse 都是 int8_tensorwise 量化,ComfyUI v0.28 以上才支援。舊版載入會報
KeyError: 'int8_tensorwise'——把 ComfyUI 更新到最新版即可解決。Text encoder 與 VAE 跟官方 Turbo 共用,不用重複下載。




