krea2-turbo-bbox

jimmycarter/krea2-turbo-bbox

A Krea 2 finetune for layout control: multi-panel comics, speech bubbles that land inside their panel, character identity held across panels, and text placed where it was asked for.

This is the final release: epoch-20-step-104548, after 20 epochs. It comes in two flavours, each as the fine-tune itself plus two EMA shadows of its weights (exponential moving averages with decay 0.9995 and 0.99975; the higher decay averages over a longer window):

• Turbo is the fine-tune merged with the base→turbo distillation delta, turbo = finetune + (krea/Krea-2-Turbo - krea/Krea-2-Raw). It samples in 8 steps with CFG off.
• Raw is the fine-tune as trained, on top of krea/Krea-2-Raw. It samples in 50 steps with CFG.

<span style="color:#e11d48">⚠️ Prompting: read this first</span>

Roughly 90% of this model's training used a compact grounding DSL, not ordinary prose. It is a plain-text layout language that places panels, characters and verbatim text in explicit [x0,y0,x1,y1] boxes on a 0–1000 grid.

→ Full spec with copy-pasteable examples: PROMPTING.md

Plain prose prompts still work. That was the other 10% of training.

<details> <summary><b>What a grounded prompt looks like</b></summary>

A four-panel comic in which an anime handywoman and an anthropomorphic tabby cat assemble a flat-pack shelf and fail completely.
@clean line art, flat colors, light cel shading; Digital illustration
~A bare apartment living room with cardboard packaging and loose screws on a pale wood floor.
p[0,0,500,500] Top-left panel: the handywoman holding up the instruction sheet, confident.
p[500,0,1000,500] Top-right panel: the cat batting a single screw off the floor.
p[0,500,500,1000] Bottom-left panel: a lopsided half-built shelf wobbling.
p[500,500,1000,1000] Bottom-right panel: the shelf collapsed, the cat sitting smugly on the wreckage.
ac:1[60,80,340,470] A young woman with short dark hair in blue overalls, holding an instruction sheet, confident grin.
fc:2[560,90,880,470] An anthropomorphic tabby cat in a tool belt, one paw raised mid-swat, innocent expression.
t[80,40,420,120]"SOME ASSEMBLY REQUIRED." black text in a rounded speech bubble
t[540,530,960,610]"IT HAS FIVE LEGS NOW." black text in a rounded speech bubble

p panel, o object, t text, pe/ac/fc person / anime / furry character. Coordinates are x first: [x0,y0,x1,y1], not [y0,x0,y1,x1]. Getting that backwards is the most common failure, and it fails silently.

</details>

Samples

Each image is one prompt with the same seed and sampler settings through four transformers: the stock model, this fine-tune, and its two EMA shadows. Grounded prompts get a second row showing the same frames with the requested boxes drawn on. Every grounded prompt is a working example of the DSL in PROMPTING.md.

---

Turbo · 8 steps, CFG off

Columns: Krea-2-Turbo · Fine-tune · EMA 0.9995 · EMA 0.99975.

Cat on a windowsill

<details><summary>prompt</summary><pre>a cat sitting on a windowsill at sunset</pre></details>

Cafe chalkboard, cursive text

<details><summary>prompt</summary><pre>a chalkboard outside a cafe that says &#x27;Best Coffee in Town&#x27; in beautiful cursive</pre></details>

Macro watch movement

<details><summary>prompt</summary><pre>extreme macro photography of a mechanical watch movement, intricate brass gears</pre></details>

Watercolour lighthouse

<details><summary>prompt</summary><pre>a watercolor painting of a lighthouse during a storm, soft bleeding colors</pre></details>

1950s diner, neon sign

<details><summary>prompt</summary><pre>A wide-angle cinematic shot of a 1950s diner at night. On the left, a large neon sign glows red with the text &#x27;MIDNIGHT DINER&#x27;. In the center foreground, a shiny chrome jukebox sits next to a vinyl booth. Through the large glass window in the background, a lonely car is parked under a flickering streetlamp. The floor is a black and white checkered pattern, and the overall lighting is moody with high contrast.</pre></details>

Circuit board, etched text

<details><summary>prompt</summary><pre>A macro top-down view of an open high-tech circuit board. In the center, a large processor chip has the word &#x27;FLUX-CORE&#x27; etched into the metal. Surrounding the chip are tiny glowing blue copper traces that look like a city grid. On the right side, several cylindrical capacitors are standing upright. To the left, a small mechanical robotic arm is mid-motion, soldering a connection with a visible tiny orange spark.</pre></details>

Meadow at golden hour

<details><summary>prompt</summary><pre>A cinematic, wide-angle shot of two people lying side-by-side in a vast, undulating meadow of lush green grass and scattered wildflowers during the golden hour. They are viewed from a high-angle, top-down perspective, their bodies forming a peaceful V shape as they stare up at a dreamlike sky painted with streaks of violet and amber. One person has their hand lightly brushing the soft clover, while the other has their eyes closed in a state of pure tranquility. The lighting is warm and ethereal, catching the fine textures of their clothing and the translucent petals of nearby daisies, while a soft-focus bokeh effect blurs the distant rolling hills into a painterly haze of emerald and gold.</pre></details>

Album cover with tracklist

<details><summary>prompt</summary><pre>A professional graphic design album cover titled &quot;SYSTEM CRITICAL&quot; in massive, bold white sans-serif letters at the top. Below the title, a sub-header reads &quot;VERSION 2.0.26&quot; in a smaller red monospace font. The middle of the cover features a detailed technical blueprint of a circuit board. Layered over the blueprint is a dense block of text containing the following tracklist in clear, legible white Arial font: &quot;1. INITIALIZE, 2. DATA STREAM, 3. CORE OVERLOAD, 4. REBOOT SEQUENCE, 5. SIGNAL LOSS, 6. YOUR MOM.&quot;. At the very bottom, in tiny but sharp fine print, the text reads: &quot;MANUFACTURED BY ARTIFICIAL INTELLIGENCE LABS GLOBAL&quot;. High resolution.</pre></details>

Grounded: coffee shop storefront

<details><summary>prompt</summary><pre>A rustic corner coffee shop photographed from across the street on an overcast morning, its green-painted storefront framed by hanging plants and a chalkboard on the sidewalk. @shallow depth of field, eye-level, 35mm lens; Photograph ~A wet cobblestone street and a blurred row of brick townhouses receding to the right. o[120,180,880,760] Green-painted storefront facade with a large display window. t[330,210,680,300]&quot;THE DAILY GRIND&quot; white serif lettering on the green fascia board o[150,640,470,880] A black sandwich-board chalkboard on the sidewalk. t[180,690,440,850]&quot;FRESH ESPRESSO\nTODAY&quot; white cursive chalk on the sandwich board pe[560,470,720,880] A barista in an apron standing in the open doorway, wiping a cup, faint smile. o[60,90,230,260] A hanging basket of trailing ivy above the window.</pre></details>

Grounded: breakfast flat-lay

<details><summary>prompt</summary><pre>An overhead macro photograph of a flat-lay breakfast spread on a dark slate board, with a small printed menu card. @crisp edge-to-edge focus, top-down, 50mm lens; Photograph ~A dark grey slate serving board with faint mineral streaks. o[110,140,470,520] A white ceramic plate with two fried eggs and avocado toast. o[520,120,820,430] A glass cup of black coffee with steam. o[540,520,900,880] A small bowl of mixed berries. o[120,600,460,900] A folded linen napkin with a brass fork resting on it. t[700,760,910,870]&quot;BRUNCH&quot; small black sans-serif text on a white menu card</pre></details>

Grounded: anime key visual

<details><summary>prompt</summary><pre>An anime character key-visual: a lone silver-haired swordswoman standing on a rooftop ledge at dusk, with a title logo in the upper-left corner. @clean line art, soft cel shading, vibrant colors; Digital illustration ~A sprawling neon-lit city skyline below a gradient orange-to-violet sky. ac:1[330,200,660,920] A young woman with long silver hair and red eyes in a black coat over a school uniform, holding a katana low at her side, neutral focused expression. o[360,430,640,880] The drawn katana catching a thin orange highlight along the blade. t[60,80,420,200]&quot;TWILIGHT EDGE&quot; bold white display logo, upper left o[0,760,1000,1000] The dark rooftop ledge the character stands on, in silhouette.</pre></details>

Grounded: noir furry detectives

<details><summary>prompt</summary><pre>A noir-styled illustration of two anthropomorphic detectives under a streetlamp in a rainy alley, with a precinct sign on the wall. @painterly rendering, cinematic contrast, fine fur texture; Digital illustration ~A narrow brick alley slick with rain, distant neon reflections on the puddles. fc:1[120,220,430,940] A grizzled grey German Shepherd in a trench coat and fedora, holding a lit cigarette, weary half-lidded eyes. fc:2[520,300,800,950] A young red fox in a police uniform holding a glowing holographic case file, attentive wide eyes. o[560,360,770,560] The blue holographic file projected above the fox&#x27;s paws. t[640,60,940,180]&quot;PRECINCT 9&quot; weathered yellow stencil lettering on the brick wall</pre></details>

Grounded: 4-panel kitchen comic

<details><summary>prompt</summary><pre>A four-panel gag comic where an anime catgirl and an anthropomorphic dog argue over a burnt cake in a tiny apartment kitchen, ending in a cloud of smoke. @clean line art, flat colors, light cel shading; Digital illustration ~A cramped pastel kitchen with a tiny oven, a wall clock, and a smoke detector. p[0,0,500,500] Top-left panel: the catgirl proudly presenting an oven. p[500,0,1000,500] Top-right panel: the dog sniffing the air, suspicious. p[0,500,500,1000] Bottom-left panel: a charred, smoking cake revealed. p[500,500,1000,1000] Bottom-right panel: both coughing in a cloud of smoke. ac:1[60,120,360,480] An anime catgirl with brown hair, cat ears, and an apron, beaming proudly. fc:2[600,120,900,480] An anthropomorphic golden retriever in a chef hat, one eyebrow raised, suspicious. t[90,40,360,110]&quot;I FOLLOWED THE RECIPE EXACTLY!&quot; black text in a round speech bubble t[560,520,960,590]&quot;OPEN A WINDOW!&quot; bold black shouting text, no bubble</pre></details>

Grounded: 3-panel summoning comic

<details><summary>prompt</summary><pre>A three-panel comic where an anime wizard apprentice and an anthropomorphic rabbit knight botch a summoning spell at a fantasy market, conjuring a chicken instead of a dragon. @bold outlines, saturated flat colors, light cel shading; Digital illustration ~A bustling cobblestone fantasy marketplace with striped stall awnings and a castle on the hill. p[0,0,1000,360] Top wide panel: the duo reading a glowing spellbook at a market stall. p[0,360,500,1000] Bottom-left panel: a chaotic magic explosion of green smoke. p[500,360,1000,1000] Bottom-right panel: a tiny confused summoned chicken stares back. ac:1[80,60,360,340] An anime wizard apprentice with purple hair and an oversized pointed hat, reading aloud, mouth wide. fc:2[620,50,940,350] An anthropomorphic white rabbit knight in shining armor holding the spellbook open, focused. o[800,560,980,940] A small bewildered chicken with a faint magic halo, summoned by mistake. t[120,30,520,100]&quot;...AND THUS I SUMMON A MIGHTY DRAGON!&quot; black text in a large speech bubble t[560,420,940,500]&quot;THAT IS A CHICKEN.&quot; flat black deadpan text, no bubble</pre></details>

Grounded: 4-panel flat-pack comic

<details><summary>prompt</summary><pre>A four-panel comic in which an anime handywoman and an anthropomorphic tabby cat assemble a flat-pack shelf and fail completely. @clean line art, flat colors, light cel shading; Digital illustration ~A bare apartment living room with cardboard packaging and loose screws on a pale wood floor. p[0,0,500,500] Top-left panel: the handywoman holding up the instruction sheet, confident. p[500,0,1000,500] Top-right panel: the cat batting a single screw off the floor. p[0,500,500,1000] Bottom-left panel: a lopsided half-built shelf wobbling. p[500,500,1000,1000] Bottom-right panel: the shelf collapsed, the cat sitting smugly on the wreckage. ac:1[60,80,340,470] A young woman with short dark hair in blue overalls, holding an instruction sheet, confident grin. fc:2[560,90,880,470] An anthropomorphic tabby cat in a tool belt, one paw raised mid-swat, innocent expression. t[80,40,420,120]&quot;SOME ASSEMBLY REQUIRED.&quot; black text in a rounded speech bubble o[80,700,440,960] The collapsed shelf, planks fanned out across the floor. t[540,530,960,610]&quot;IT HAS FIVE LEGS NOW.&quot; black text in a rounded speech bubble</pre></details>

Grounded: 3-panel strip, one character

<details><summary>prompt</summary><pre>A three-panel vertical comic strip following one barista through an escalating espresso order across a single shift. @soft line art, flat colors with light gradients; Digital illustration ~A small café counter with a brass espresso machine and a shelf of mugs. p[0,0,1000,333] Top panel: the barista takes an order, alert and cheerful. p[0,333,1000,666] Middle panel: the barista pulling shots, sleeves pushed up, focused. p[0,666,1000,1000] Bottom panel: the barista slumped against the counter at closing time. ac:1[80,40,400,320] A barista with a messy green hoodie under an apron and short auburn hair, smiling brightly. t[450,30,950,110]&quot;ONE ESPRESSO, PLEASE.&quot; black text in a speech bubble ac:1[560,370,900,650] The same barista mid-shift, sleeves rolled, brow furrowed at the machine. t[60,360,520,440]&quot;...MAKE IT SIX.&quot; black text in a speech bubble ac:1[100,700,420,980] The same barista slumped over the counter, hollow-eyed, apron askew. o[600,700,940,980] A precarious pyramid of empty espresso cups. t[520,690,960,780]&quot;SLEEP IS A CONSTRUCT.&quot; black text in a jagged speech bubble</pre></details>

Grounded: 2-panel wide, SFX text

<details><summary>prompt</summary><pre>A two-panel wide comic where an armoured knight takes a swing at a birthday piñata and destroys it utterly. @heavy outlines, saturated flat colors, speed lines; Digital illustration ~A suburban back garden strung with paper bunting and folding chairs. p[0,0,500,1000] Left panel: the knight winding up an enormous two-handed swing. p[500,0,1000,1000] Right panel: the piñata detonating in a shower of sweets. pe:1[80,200,420,900] A knight in full plate armour swinging a wooden stick like a greatsword, visor down. o[560,150,980,700] The bursting piñata, showering wrapped sweets in every direction. t[540,120,960,320]&quot;KRAK!!&quot; huge jagged yellow SFX lettering, no bubble t[60,820,440,920]&quot;FINALLY.&quot; small flat black deadpan caption, no bubble</pre></details>

Grounded: splash page + 3 panels

<details><summary>prompt</summary><pre>A comic page with a wide splash panel of a failing brass airship over a canyon, and three narrow panels beneath it. @inked line art with cross-hatching, muted flat colors; Digital illustration ~A vast red-rock canyon with layered mesas fading into haze. p[0,0,1000,400] Wide top splash panel: the airship listing badly over the canyon. p[0,400,333,1000] Bottom-left panel: the pilot yanking a jammed lever in the gondola. p[333,400,666,1000] Bottom-centre panel: a gauge needle buried in the red. p[666,400,1000,1000] Bottom-right panel: the pilot shrugging, resigned, mail sack in paw. fc:1[120,60,420,360] An otter pilot in a leather cap and brass goggles leaning out of the gondola. o[520,80,940,340] The brass airship listing to one side, one propeller trailing smoke. t[60,30,520,120]&quot;THE LAST MAIL RUN&quot; bold cream display lettering, upper left of the splash t[20,430,310,520]&quot;ALTITUDE?&quot; black text in a small speech bubble t[690,430,980,530]&quot;OPTIONAL.&quot; black text in a small speech bubble</pre></details>

Grounded: anime key visual + logo

<details><summary>prompt</summary><pre>An anime key visual of a crimson-cloaked archer drawing a longbow on a windswept hillside at dusk, with a title logo and subtitle in the upper left. @clean line art, soft cel shading, high saturation; Digital illustration ~A long grassy ridge under a banded orange and deep-blue sky with distant birds. ac:1[300,180,700,940] A young archer in a crimson cloak over dark leather, black hair whipping sideways, one eye narrowed along the arrow. o[260,420,560,900] The fully drawn longbow, string taut, catching an amber highlight. t[80,70,520,190]&quot;SCARLET VOLLEY&quot; bold cream display logo with a thin gold outline t[90,210,470,280]&quot;SEASON TWO — THIS AUTUMN&quot; small spaced white sans-serif subtitle o[0,820,1000,1000] Foreground grass blades in near-silhouette across the bottom edge.</pre></details>

Grounded: three-piece furry band

<details><summary>prompt</summary><pre>Three anthropomorphic musicians mid-set on a cramped club stage beneath a hand-painted band banner. @painterly rendering with visible brush texture, fine fur detail; Digital illustration ~A small club stage with taped cables, a brick back wall, and a haze of stage smoke. fc:1[60,250,330,950] A stocky badger in a sleeveless denim vest playing a battered bass guitar, eyes shut. fc:2[360,180,660,960] A lynx vocalist leaning into the microphone, tufted ears back, mouth open mid-note. fc:3[690,280,970,900] A lean hare behind the kit, sticks blurred mid-strike, ears pinned by motion. o[330,600,700,900] A battered drum kit with mismatched cymbals. t[220,60,780,180]&quot;THE NIGHT SHIFT&quot; hand-painted cream lettering on a canvas banner behind the band</pre></details>

Grounded: screen-printed poster

<details><summary>prompt</summary><pre>A screen-printed event poster for an autumn outdoor festival, with a large headline, a date line, and a strip of fine print at the bottom. @woodcut-style illustration, three-colour screen print, coarse paper grain; Poster ~A cream paper ground with a faint halftone texture and a thin rule border. t[100,80,900,220]&quot;HARVEST LIGHT&quot; huge deep-red condensed slab-serif headline t[180,250,820,330]&quot;OCTOBER 14 · OLD MILL YARD&quot; medium dark-green spaced sans-serif date line o[150,380,850,830] A woodcut-style illustration of a lantern-strung courtyard with long tables. t[150,900,850,960]&quot;DOORS 7PM · ALL AGES · BRING A COAT&quot; tiny black fine print in a single centred line</pre></details>

Grounded: ramen shop at night

<details><summary>prompt</summary><pre>A tiny ramen shop photographed from the wet street at night, its lit counter glowing behind a noren curtain. @shallow depth of field, eye-level, 35mm lens; Photograph ~A narrow alley of wet asphalt with smeared neon reflections receding into darkness. o[80,150,520,850] The ramen shop front: a timber facade with a noren curtain over a lit counter. t[150,200,470,300]&quot;RAMEN&quot; glowing red sign lettering above the shopfront pe:1[180,400,340,820] A chef in a white cap and towel headband leaning through the shop window. o[560,560,880,900] A chalkboard menu propped against the doorframe. t[590,600,860,860]&quot;TONKOTSU\nMISO\nSHOYU&quot; white chalk lettering, three stacked lines</pre></details>

Grounded: coffee bag product shot

<details><summary>prompt</summary><pre>A studio product photograph of a matte black coffee bag standing on a pale stone plinth with scattered beans. @crisp focus on the bag face, slight depth falloff, 85mm lens; Photograph ~A seamless warm-grey studio backdrop with a soft vertical gradient. o[300,250,700,850] A matte black coffee bag standing upright, top edge crimped. t[360,400,640,520]&quot;SINGLE ORIGIN&quot; white spaced sans-serif lettering across the bag face t[400,560,600,620]&quot;ETHIOPIA · 250g&quot; small grey sans-serif line beneath the main lettering o[120,700,880,900] The pale stone plinth surface with a few scattered coffee beans.</pre></details>

Retro travel poster

<details><summary>prompt</summary><pre>A 1960s screen-printed travel poster for a fictional mountain railway. A stylised red funicular climbs a steep green slope on the right. Large cream letters across the top read &#x27;ALPENBAHN&#x27;. Beneath the illustration a single line of smaller black type reads &#x27;DEPARTURES HOURLY FROM THE VALLEY STATION&#x27;. Limited four-colour palette, visible ink texture, coarse paper grain, flat graphic shapes with no gradients.</pre></details>

Coastal highway at dawn

<details><summary>prompt</summary><pre>A cinematic wide shot of an empty coastal highway at dawn, shot from a low tripod on the centre line. Wet tarmac reflects a pale pink and grey sky, telephone poles recede toward a distant headland, and a thin band of sea mist hangs over the guardrail on the right. Anamorphic lens character with gentle flare, deep focus, muted desaturated colour grade, no people and no text anywhere in the frame.</pre></details>

Ink and watercolour tram

<details><summary>prompt</summary><pre>A loose ink-and-watercolour sketch of a Lisbon tram climbing a narrow tiled street. Confident dark ink linework describes the tram and the balconies above it, with washes of ochre, teal and dusty rose bleeding softly outside the lines. Large areas of untouched paper are left white, and the pigment pools and granulates where the washes overlap.</pre></details>

---

Raw · 50 steps, CFG 5.5

Columns: Krea-2-Raw · Fine-tune · EMA 0.9995 · EMA 0.99975.

Cat on a windowsill

<details><summary>prompt</summary><pre>a cat sitting on a windowsill at sunset</pre></details>

Cafe chalkboard, cursive text

<details><summary>prompt</summary><pre>a chalkboard outside a cafe that says &#x27;Best Coffee in Town&#x27; in beautiful cursive</pre></details>

Macro watch movement

<details><summary>prompt</summary><pre>extreme macro photography of a mechanical watch movement, intricate brass gears</pre></details>

Watercolour lighthouse

<details><summary>prompt</summary><pre>a watercolor painting of a lighthouse during a storm, soft bleeding colors</pre></details>

1950s diner, neon sign

<details><summary>prompt</summary><pre>A wide-angle cinematic shot of a 1950s diner at night. On the left, a large neon sign glows red with the text &#x27;MIDNIGHT DINER&#x27;. In the center foreground, a shiny chrome jukebox sits next to a vinyl booth. Through the large glass window in the background, a lonely car is parked under a flickering streetlamp. The floor is a black and white checkered pattern, and the overall lighting is moody with high contrast.</pre></details>

Circuit board, etched text

<details><summary>prompt</summary><pre>A macro top-down view of an open high-tech circuit board. In the center, a large processor chip has the word &#x27;FLUX-CORE&#x27; etched into the metal. Surrounding the chip are tiny glowing blue copper traces that look like a city grid. On the right side, several cylindrical capacitors are standing upright. To the left, a small mechanical robotic arm is mid-motion, soldering a connection with a visible tiny orange spark.</pre></details>

Meadow at golden hour

<details><summary>prompt</summary><pre>A cinematic, wide-angle shot of two people lying side-by-side in a vast, undulating meadow of lush green grass and scattered wildflowers during the golden hour. They are viewed from a high-angle, top-down perspective, their bodies forming a peaceful V shape as they stare up at a dreamlike sky painted with streaks of violet and amber. One person has their hand lightly brushing the soft clover, while the other has their eyes closed in a state of pure tranquility. The lighting is warm and ethereal, catching the fine textures of their clothing and the translucent petals of nearby daisies, while a soft-focus bokeh effect blurs the distant rolling hills into a painterly haze of emerald and gold.</pre></details>

Album cover with tracklist

<details><summary>prompt</summary><pre>A professional graphic design album cover titled &quot;SYSTEM CRITICAL&quot; in massive, bold white sans-serif letters at the top. Below the title, a sub-header reads &quot;VERSION 2.0.26&quot; in a smaller red monospace font. The middle of the cover features a detailed technical blueprint of a circuit board. Layered over the blueprint is a dense block of text containing the following tracklist in clear, legible white Arial font: &quot;1. INITIALIZE, 2. DATA STREAM, 3. CORE OVERLOAD, 4. REBOOT SEQUENCE, 5. SIGNAL LOSS, 6. YOUR MOM.&quot;. At the very bottom, in tiny but sharp fine print, the text reads: &quot;MANUFACTURED BY ARTIFICIAL INTELLIGENCE LABS GLOBAL&quot;. High resolution.</pre></details>

Grounded: coffee shop storefront

<details><summary>prompt</summary><pre>A rustic corner coffee shop photographed from across the street on an overcast morning, its green-painted storefront framed by hanging plants and a chalkboard on the sidewalk. @shallow depth of field, eye-level, 35mm lens; Photograph ~A wet cobblestone street and a blurred row of brick townhouses receding to the right. o[120,180,880,760] Green-painted storefront facade with a large display window. t[330,210,680,300]&quot;THE DAILY GRIND&quot; white serif lettering on the green fascia board o[150,640,470,880] A black sandwich-board chalkboard on the sidewalk. t[180,690,440,850]&quot;FRESH ESPRESSO\nTODAY&quot; white cursive chalk on the sandwich board pe[560,470,720,880] A barista in an apron standing in the open doorway, wiping a cup, faint smile. o[60,90,230,260] A hanging basket of trailing ivy above the window.</pre></details>

Grounded: breakfast flat-lay

<details><summary>prompt</summary><pre>An overhead macro photograph of a flat-lay breakfast spread on a dark slate board, with a small printed menu card. @crisp edge-to-edge focus, top-down, 50mm lens; Photograph ~A dark grey slate serving board with faint mineral streaks. o[110,140,470,520] A white ceramic plate with two fried eggs and avocado toast. o[520,120,820,430] A glass cup of black coffee with steam. o[540,520,900,880] A small bowl of mixed berries. o[120,600,460,900] A folded linen napkin with a brass fork resting on it. t[700,760,910,870]&quot;BRUNCH&quot; small black sans-serif text on a white menu card</pre></details>

Grounded: anime key visual

<details><summary>prompt</summary><pre>An anime character key-visual: a lone silver-haired swordswoman standing on a rooftop ledge at dusk, with a title logo in the upper-left corner. @clean line art, soft cel shading, vibrant colors; Digital illustration ~A sprawling neon-lit city skyline below a gradient orange-to-violet sky. ac:1[330,200,660,920] A young woman with long silver hair and red eyes in a black coat over a school uniform, holding a katana low at her side, neutral focused expression. o[360,430,640,880] The drawn katana catching a thin orange highlight along the blade. t[60,80,420,200]&quot;TWILIGHT EDGE&quot; bold white display logo, upper left o[0,760,1000,1000] The dark rooftop ledge the character stands on, in silhouette.</pre></details>

Grounded: noir furry detectives

<details><summary>prompt</summary><pre>A noir-styled illustration of two anthropomorphic detectives under a streetlamp in a rainy alley, with a precinct sign on the wall. @painterly rendering, cinematic contrast, fine fur texture; Digital illustration ~A narrow brick alley slick with rain, distant neon reflections on the puddles. fc:1[120,220,430,940] A grizzled grey German Shepherd in a trench coat and fedora, holding a lit cigarette, weary half-lidded eyes. fc:2[520,300,800,950] A young red fox in a police uniform holding a glowing holographic case file, attentive wide eyes. o[560,360,770,560] The blue holographic file projected above the fox&#x27;s paws. t[640,60,940,180]&quot;PRECINCT 9&quot; weathered yellow stencil lettering on the brick wall</pre></details>

Grounded: 4-panel kitchen comic

<details><summary>prompt</summary><pre>A four-panel gag comic where an anime catgirl and an anthropomorphic dog argue over a burnt cake in a tiny apartment kitchen, ending in a cloud of smoke. @clean line art, flat colors, light cel shading; Digital illustration ~A cramped pastel kitchen with a tiny oven, a wall clock, and a smoke detector. p[0,0,500,500] Top-left panel: the catgirl proudly presenting an oven. p[500,0,1000,500] Top-right panel: the dog sniffing the air, suspicious. p[0,500,500,1000] Bottom-left panel: a charred, smoking cake revealed. p[500,500,1000,1000] Bottom-right panel: both coughing in a cloud of smoke. ac:1[60,120,360,480] An anime catgirl with brown hair, cat ears, and an apron, beaming proudly. fc:2[600,120,900,480] An anthropomorphic golden retriever in a chef hat, one eyebrow raised, suspicious. t[90,40,360,110]&quot;I FOLLOWED THE RECIPE EXACTLY!&quot; black text in a round speech bubble t[560,520,960,590]&quot;OPEN A WINDOW!&quot; bold black shouting text, no bubble</pre></details>

Grounded: 3-panel summoning comic

<details><summary>prompt</summary><pre>A three-panel comic where an anime wizard apprentice and an anthropomorphic rabbit knight botch a summoning spell at a fantasy market, conjuring a chicken instead of a dragon. @bold outlines, saturated flat colors, light cel shading; Digital illustration ~A bustling cobblestone fantasy marketplace with striped stall awnings and a castle on the hill. p[0,0,1000,360] Top wide panel: the duo reading a glowing spellbook at a market stall. p[0,360,500,1000] Bottom-left panel: a chaotic magic explosion of green smoke. p[500,360,1000,1000] Bottom-right panel: a tiny confused summoned chicken stares back. ac:1[80,60,360,340] An anime wizard apprentice with purple hair and an oversized pointed hat, reading aloud, mouth wide. fc:2[620,50,940,350] An anthropomorphic white rabbit knight in shining armor holding the spellbook open, focused. o[800,560,980,940] A small bewildered chicken with a faint magic halo, summoned by mistake. t[120,30,520,100]&quot;...AND THUS I SUMMON A MIGHTY DRAGON!&quot; black text in a large speech bubble t[560,420,940,500]&quot;THAT IS A CHICKEN.&quot; flat black deadpan text, no bubble</pre></details>

Grounded: 4-panel flat-pack comic

<details><summary>prompt</summary><pre>A four-panel comic in which an anime handywoman and an anthropomorphic tabby cat assemble a flat-pack shelf and fail completely. @clean line art, flat colors, light cel shading; Digital illustration ~A bare apartment living room with cardboard packaging and loose screws on a pale wood floor. p[0,0,500,500] Top-left panel: the handywoman holding up the instruction sheet, confident. p[500,0,1000,500] Top-right panel: the cat batting a single screw off the floor. p[0,500,500,1000] Bottom-left panel: a lopsided half-built shelf wobbling. p[500,500,1000,1000] Bottom-right panel: the shelf collapsed, the cat sitting smugly on the wreckage. ac:1[60,80,340,470] A young woman with short dark hair in blue overalls, holding an instruction sheet, confident grin. fc:2[560,90,880,470] An anthropomorphic tabby cat in a tool belt, one paw raised mid-swat, innocent expression. t[80,40,420,120]&quot;SOME ASSEMBLY REQUIRED.&quot; black text in a rounded speech bubble o[80,700,440,960] The collapsed shelf, planks fanned out across the floor. t[540,530,960,610]&quot;IT HAS FIVE LEGS NOW.&quot; black text in a rounded speech bubble</pre></details>

Grounded: 3-panel strip, one character

<details><summary>prompt</summary><pre>A three-panel vertical comic strip following one barista through an escalating espresso order across a single shift. @soft line art, flat colors with light gradients; Digital illustration ~A small café counter with a brass espresso machine and a shelf of mugs. p[0,0,1000,333] Top panel: the barista takes an order, alert and cheerful. p[0,333,1000,666] Middle panel: the barista pulling shots, sleeves pushed up, focused. p[0,666,1000,1000] Bottom panel: the barista slumped against the counter at closing time. ac:1[80,40,400,320] A barista with a messy green hoodie under an apron and short auburn hair, smiling brightly. t[450,30,950,110]&quot;ONE ESPRESSO, PLEASE.&quot; black text in a speech bubble ac:1[560,370,900,650] The same barista mid-shift, sleeves rolled, brow furrowed at the machine. t[60,360,520,440]&quot;...MAKE IT SIX.&quot; black text in a speech bubble ac:1[100,700,420,980] The same barista slumped over the counter, hollow-eyed, apron askew. o[600,700,940,980] A precarious pyramid of empty espresso cups. t[520,690,960,780]&quot;SLEEP IS A CONSTRUCT.&quot; black text in a jagged speech bubble</pre></details>

Grounded: 2-panel wide, SFX text

<details><summary>prompt</summary><pre>A two-panel wide comic where an armoured knight takes a swing at a birthday piñata and destroys it utterly. @heavy outlines, saturated flat colors, speed lines; Digital illustration ~A suburban back garden strung with paper bunting and folding chairs. p[0,0,500,1000] Left panel: the knight winding up an enormous two-handed swing. p[500,0,1000,1000] Right panel: the piñata detonating in a shower of sweets. pe:1[80,200,420,900] A knight in full plate armour swinging a wooden stick like a greatsword, visor down. o[560,150,980,700] The bursting piñata, showering wrapped sweets in every direction. t[540,120,960,320]&quot;KRAK!!&quot; huge jagged yellow SFX lettering, no bubble t[60,820,440,920]&quot;FINALLY.&quot; small flat black deadpan caption, no bubble</pre></details>

Grounded: splash page + 3 panels

<details><summary>prompt</summary><pre>A comic page with a wide splash panel of a failing brass airship over a canyon, and three narrow panels beneath it. @inked line art with cross-hatching, muted flat colors; Digital illustration ~A vast red-rock canyon with layered mesas fading into haze. p[0,0,1000,400] Wide top splash panel: the airship listing badly over the canyon. p[0,400,333,1000] Bottom-left panel: the pilot yanking a jammed lever in the gondola. p[333,400,666,1000] Bottom-centre panel: a gauge needle buried in the red. p[666,400,1000,1000] Bottom-right panel: the pilot shrugging, resigned, mail sack in paw. fc:1[120,60,420,360] An otter pilot in a leather cap and brass goggles leaning out of the gondola. o[520,80,940,340] The brass airship listing to one side, one propeller trailing smoke. t[60,30,520,120]&quot;THE LAST MAIL RUN&quot; bold cream display lettering, upper left of the splash t[20,430,310,520]&quot;ALTITUDE?&quot; black text in a small speech bubble t[690,430,980,530]&quot;OPTIONAL.&quot; black text in a small speech bubble</pre></details>

Grounded: anime key visual + logo

<details><summary>prompt</summary><pre>An anime key visual of a crimson-cloaked archer drawing a longbow on a windswept hillside at dusk, with a title logo and subtitle in the upper left. @clean line art, soft cel shading, high saturation; Digital illustration ~A long grassy ridge under a banded orange and deep-blue sky with distant birds. ac:1[300,180,700,940] A young archer in a crimson cloak over dark leather, black hair whipping sideways, one eye narrowed along the arrow. o[260,420,560,900] The fully drawn longbow, string taut, catching an amber highlight. t[80,70,520,190]&quot;SCARLET VOLLEY&quot; bold cream display logo with a thin gold outline t[90,210,470,280]&quot;SEASON TWO — THIS AUTUMN&quot; small spaced white sans-serif subtitle o[0,820,1000,1000] Foreground grass blades in near-silhouette across the bottom edge.</pre></details>

Grounded: three-piece furry band

<details><summary>prompt</summary><pre>Three anthropomorphic musicians mid-set on a cramped club stage beneath a hand-painted band banner. @painterly rendering with visible brush texture, fine fur detail; Digital illustration ~A small club stage with taped cables, a brick back wall, and a haze of stage smoke. fc:1[60,250,330,950] A stocky badger in a sleeveless denim vest playing a battered bass guitar, eyes shut. fc:2[360,180,660,960] A lynx vocalist leaning into the microphone, tufted ears back, mouth open mid-note. fc:3[690,280,970,900] A lean hare behind the kit, sticks blurred mid-strike, ears pinned by motion. o[330,600,700,900] A battered drum kit with mismatched cymbals. t[220,60,780,180]&quot;THE NIGHT SHIFT&quot; hand-painted cream lettering on a canvas banner behind the band</pre></details>

Grounded: screen-printed poster

<details><summary>prompt</summary><pre>A screen-printed event poster for an autumn outdoor festival, with a large headline, a date line, and a strip of fine print at the bottom. @woodcut-style illustration, three-colour screen print, coarse paper grain; Poster ~A cream paper ground with a faint halftone texture and a thin rule border. t[100,80,900,220]&quot;HARVEST LIGHT&quot; huge deep-red condensed slab-serif headline t[180,250,820,330]&quot;OCTOBER 14 · OLD MILL YARD&quot; medium dark-green spaced sans-serif date line o[150,380,850,830] A woodcut-style illustration of a lantern-strung courtyard with long tables. t[150,900,850,960]&quot;DOORS 7PM · ALL AGES · BRING A COAT&quot; tiny black fine print in a single centred line</pre></details>

Grounded: ramen shop at night

<details><summary>prompt</summary><pre>A tiny ramen shop photographed from the wet street at night, its lit counter glowing behind a noren curtain. @shallow depth of field, eye-level, 35mm lens; Photograph ~A narrow alley of wet asphalt with smeared neon reflections receding into darkness. o[80,150,520,850] The ramen shop front: a timber facade with a noren curtain over a lit counter. t[150,200,470,300]&quot;RAMEN&quot; glowing red sign lettering above the shopfront pe:1[180,400,340,820] A chef in a white cap and towel headband leaning through the shop window. o[560,560,880,900] A chalkboard menu propped against the doorframe. t[590,600,860,860]&quot;TONKOTSU\nMISO\nSHOYU&quot; white chalk lettering, three stacked lines</pre></details>

Grounded: coffee bag product shot

<details><summary>prompt</summary><pre>A studio product photograph of a matte black coffee bag standing on a pale stone plinth with scattered beans. @crisp focus on the bag face, slight depth falloff, 85mm lens; Photograph ~A seamless warm-grey studio backdrop with a soft vertical gradient. o[300,250,700,850] A matte black coffee bag standing upright, top edge crimped. t[360,400,640,520]&quot;SINGLE ORIGIN&quot; white spaced sans-serif lettering across the bag face t[400,560,600,620]&quot;ETHIOPIA · 250g&quot; small grey sans-serif line beneath the main lettering o[120,700,880,900] The pale stone plinth surface with a few scattered coffee beans.</pre></details>

Retro travel poster

<details><summary>prompt</summary><pre>A 1960s screen-printed travel poster for a fictional mountain railway. A stylised red funicular climbs a steep green slope on the right. Large cream letters across the top read &#x27;ALPENBAHN&#x27;. Beneath the illustration a single line of smaller black type reads &#x27;DEPARTURES HOURLY FROM THE VALLEY STATION&#x27;. Limited four-colour palette, visible ink texture, coarse paper grain, flat graphic shapes with no gradients.</pre></details>

Coastal highway at dawn

<details><summary>prompt</summary><pre>A cinematic wide shot of an empty coastal highway at dawn, shot from a low tripod on the centre line. Wet tarmac reflects a pale pink and grey sky, telephone poles recede toward a distant headland, and a thin band of sea mist hangs over the guardrail on the right. Anamorphic lens character with gentle flare, deep focus, muted desaturated colour grade, no people and no text anywhere in the frame.</pre></details>

Ink and watercolour tram

<details><summary>prompt</summary><pre>A loose ink-and-watercolour sketch of a Lisbon tram climbing a narrow tiled street. Confident dark ink linework describes the tram and the balconies above it, with washes of ochre, teal and dusty rose bleeding softly outside the lines. Large areas of untouched paper are left white, and the pigment pools and granulates where the washes overlap.</pre></details>

Usage

This repo holds only transformers. Everything else (VAE, text encoder, tokenizer, scheduler) is identical to stock Krea 2, so load a transformer from its subfolder into a stock pipeline. Any row of the table above works the same way, e.g. subfolder="turbo/ema_0.9995/transformer".

diffusers

import torch
from diffusers import Krea2Pipeline, Krea2Transformer2DModel

# Turbo: 8 steps, CFG off. is_distilled=True selects the fixed timestep shift (mu=1.15) that
# distilled weights expect.
tf = Krea2Transformer2DModel.from_pretrained(
    "jimmycarter/krea2-turbo-bbox", subfolder="turbo/transformer", torch_dtype=torch.bfloat16)
pipe = Krea2Pipeline.from_pretrained(
    "krea/Krea-2-Raw", transformer=tf, is_distilled=True, torch_dtype=torch.bfloat16)
pipe.to("cuda")
image = pipe("a photo of a cat", num_inference_steps=8, guidance_scale=0.0).images[0]

import torch
from diffusers import Krea2Pipeline, Krea2Transformer2DModel

# Raw: 50 steps with CFG. diffusers' guidance_scale is the CFG scale minus one: 4.5 is CFG 5.5.
tf = Krea2Transformer2DModel.from_pretrained(
    "jimmycarter/krea2-turbo-bbox", subfolder="raw/transformer", torch_dtype=torch.bfloat16)
pipe = Krea2Pipeline.from_pretrained("krea/Krea-2-Raw", transformer=tf, torch_dtype=torch.bfloat16)
pipe.to("cuda")
image = pipe("a photo of a cat", num_inference_steps=50, guidance_scale=4.5).images[0]

ComfyUI

Every variant as a single file in Krea's native layout, which is what ComfyUI loads:

Put them in ComfyUI/models/diffusion_models/. The text encoder and VAE are unchanged from stock Krea 2, so take them from Comfy-Org/Krea-2: text_encoders/qwen3vl_4b_fp8_scaled.safetensors and vae/qwen_image_vae.safetensors. The built-in Krea 2 text-to-image templates work as-is with the matching file swapped in for the stock diffusion model.

LoRAs

Lighter, approximate versions of the fine-tune and its EMAs, as LoRAs against krea/Krea-2-Raw: per Linear layer, the truncated SVD of (fine-tune − Raw). They are partial. This was a full-rank fine-tune whose change is spread over almost every direction of every weight matrix, so even rank 256 holds only a small part of it, and a LoRA can't carry the norms, biases or modulation tables at all. Measured on held-out training samples, the fine-tune's LoRA reproduces about 12% / 18% / 28% of the full fine-tune's effect on the model's predictions at rank 64 / 128 / 256, and 16% / 24% / 38% on the high-noise steps that set the layout. For the behaviour shown above, use the full weights; the LoRAs are for setups that can only take a LoRA.

They can also be applied to krea/Krea-2-Turbo. The turbo weights above are finetune + (Turbo − Raw), which is Turbo + (finetune − Raw), the change these LoRAs approximate; their effect on Turbo wasn't measured separately. Sample on Turbo at 8 steps with CFG off, on Raw at 50 steps with CFG 5.5.

import torch
from diffusers import Krea2Pipeline

pipe = Krea2Pipeline.from_pretrained(
    "krea/Krea-2-Turbo", is_distilled=True, torch_dtype=torch.bfloat16)
pipe.load_lora_weights("jimmycarter/krea2-turbo-bbox", weight_name="lora/krea2-bbox-lora-r256.safetensors")
pipe.to("cuda")
image = pipe("a photo of a cat", num_inference_steps=8, guidance_scale=0.0).images[0]

In ComfyUI, put a file in ComfyUI/models/loras/ and apply it at strength 1.0 with LoraLoaderModelOnly to the stock Krea 2 diffusion model, Raw or Turbo.

Training checkpoints

training/ holds the per-epoch turbo merges published while training ran (epochs 1, 2, 3, 4, 5, 11, 12, 13, 14, 15, 16, 17, 18, 19). They are kept for reference and superseded by the final weights above. Load one with subfolder="training/<epoch>/transformer", sampled like the turbo weights.

Training data

The initial pool of 30m images from the pre-AI internet was k-means clustered using DINOv3 and SigLIP2 and aesthetics ranking. Any images containing drawn art along with comics were further sampled with both wd-tagger and JTP, and re-ranked for diversity. Final dataset of images was 20,000 general images (photos etc), 10,000 comics, and 10,000 art images.

The 40,000 images were bboxed with a custom labelling pipeline using google/gemma-4-31B-it and nvidia/LocateAnything-3B. After the JSON output was completed the prompts were compressed to 512 tokens maximum according to PROMPTING.md and used for training.