Redrob Image

My role

Model selection (Z-Image Turbo open weights), blind preference design, and Studio bake-off ops shared with the client shell.

Purpose

Ship an Apache-2.0 open-weight image model fast enough for interactive bake-offs and competitive enough on preference against a strong hosted baseline, with weights free to use and free for commercial use, and light enough that one consumer GPU is the target rather than the fallback.

Decisions

• Drop Stable Diffusion for natural-prompt anatomy

  Tested Stable Diffusion against Klein 4B and Z-Image Turbo. SD requires a narrow prompt dialect for clean anatomy; ordinary language failed. That dialect was not acceptable.

• Reject Klein despite photorealism

  Klein 4B was strong on skin and light but broke typography on graphic and print-adjacent work. Photorealism alone was insufficient.

• Ship open weights on Z-Image Turbo

  Released weights are a Z-Image Turbo merge (Diffusers transformer or single ComfyUI UNET) at about 8 DiT steps. Tuned for more realistic skin, light, and texture than the bare base. Apache-2.0 on the weights themselves, not only the code: free to use, free for commercial use, redistributable.

• No custom runtime to learn

  It loads through the standard Diffusers pipeline API with from_pretrained, and ships to ComfyUI as one merged UNET. There is no wrapper of ours in the path, nothing to sign up for, and no key to fetch.

• Turbo sampling costs you the negative prompt

  Eight steps only works without classifier-free guidance (guidance_scale=0, ComfyUI cfg 1), and with guidance off the negative prompt is ignored entirely. Avoidances have to be written into the positive prompt, which is a real constraint rather than a documentation gap.

• Keep strengths explicit

  Strong on photo, portrait, and mood; weak on legible text, including Hangul and most non-Latin script, and well below the photo results on typography-heavy graphic and print work. Preference study pattern matches that split. Text-bearing surfaces keep a frontier-API escape hatch.

Challenges

• Hosted APIs were too slow to iterate

  At 20s-3min per image, LoRA and checkpoint sweeps were impractical. Self-hosted ~1.8s end-to-end (same hardware conditions as above) made interactive bake-offs feasible.

• Category cells were underpowered

  Six comparisons per category cannot establish per-style winners. Shipping used the overall n=90 preference result only. Graphic and print cells failed in the same direction.

• Limits

  n=90, internal raters, one prompt distribution. Underpowered by design; raters not blind to who built the system under test.

Impact

• Measured path with open weights

  44-16-30 justified continued development, not a parity claim. Latency supported interactive bake-offs. Unit cost ~1/40 of API list price at measured 50% utilization (owned hardware vs published price; moves with utilization). Weights are Apache-2.0 and free for commercial use.

• Distillation, not a proprietary stack

  Speed came from base-model distillation on Z-Image Turbo, not a closed serving stack. The evaluation evidence and open weights are the durable assets.

Unit cost

Fully loaded self-hosted cost per image vs GPT Image 2 at $0.04. ~1/40 of API list price at measured 50% GPU utilization. Owned-hardware cost vs published price; not like-for-like; utilization-dependent.

Self-hosted unit cost is dominated by fixed GPU capacity. The figure is undefined without a utilization assumption.

Outputs

• [Medium and Style] Surreal fine-art editorial photography with analog film aesthetics, heavy grain, soft halation, and poster-like color blocking. Vertical full-body composition, centered and nearly symmetrical, photographed from a slightly low eye-level angle with a normal-to-short-telephoto lens. Strongly stylized red-and-black palette with cyan-blue clothing, resembling an experimental album cover or vintage screen-printed photograph. [Subject(s) and Clothing] A solitary adult human figure stands centered in the frame. The person has a shaved or closely cropped head, but the face and scalp are completely overexposed into a smooth, featureless white silhouette. They wear an oversized pale cyan-blue pullover jacket with a high raised collar, broad dropped shoulders, voluminous sleeves, gathered elastic cuffs, and a loose bloused waist. A subtle curved seam or pocket-like fold appears on the upper right side of the chest. The lower body is covered by very dark, nearly black, loose straight-leg trousers. Both hands are visible as dark silhouettes emerging from the sleeves. No accessories or footwear details are clearly visible. [Pose and Expression] The figure faces directly toward the camera and stands upright with squared shoulders. Both arms hang naturally at the sides, with the hands slightly separated from the body. One leg appears subtly advanced, suggesting a slow forward step. The head is held straight. No facial expression is visible because the face is entirely erased by intense white overexposure. [Background and Environment] A barren, rocky outdoor landscape stretches across the lower half of the image. The foreground is almost completely black, containing irregular stones, low scrub, and sparse branching plants outlined by red-orange rim light. Larger dark rock formations sit behind and beside the figure, including a prominent jagged rock to the figure's lower right. The distant horizon is obscured by red haze, with several soft, rounded dark silhouettes resembling boulders or low vegetation. The upper two-thirds of the frame is filled by a flat, intensely saturated orange-red sky with subtle mottled texture and film grain. [Lighting, Atmosphere and Colors] Extreme frontal or backlit overexposure turns the head into pure ivory-white light and creates a glowing orange-red halo around the shoulders and outer silhouette. The jacket transitions from pale white-blue near the collar and shoulders into deeper cyan and muted blue toward the torso and sleeves. The lower body and ground remain almost entirely black, producing stark tonal separation. Dominant colors are vermilion, scarlet, burnt orange, cyan-blue, ivory white, and deep black. The atmosphere is apocalyptic, alien, ominous, and cinematic, with dense red haze, strong contrast, visible grain, slight color bleed, and soft glowing edges.
• film grain, faded color tones, low contrast, long shadows, artistic framing, almost cinematic feel, editorial photography, moody photography - analog black-and-white portrait of h4sr3t, her head tilted downward and to one side, eyes closed, lips painted a deep matte shade. She wears a fitted ribbed light beige turtleneck and a pinstriped high-waisted skirt, seated sideways on a stone rooftop ledge with her torso angled toward the camera and one shoulder dropped lower than the other. Her right hand is raised beside her temple, two fingers loosely holding a lit cigarette close to her fringe, thin curls of smoke drifting up and tangling into her hair, while her left arm reaches back and down, palm braced flat against the ledge behind her hip for support. Behind her sprawls a rooftop skyline of mansard buildings, and further back the dark, sharply defined silhouette of the Eiffel Tower cuts against the pale, flat sky. Soft overcast light wraps around her features while grain and subtle contrast deepen the folds of her sweater and the texture of her hair, lending the image a hushed, detached elegance, wholly analog and cinematic in tone.
• A high-contrast surrealist painting of a brilliant white desert under a cloudless, searing midday sun. The landscape is dominated by impossibly long, pitch-black shadows that possess their own life and form, constructing entire cities across the dunes. These shadow-cities feature towering skyscrapers, marketplaces teeming with shadowy figures, and swaying shadow trees that exist without physical trunks. Shadow birds fly through the empty air above the sand. In stark contrast, a few "real" objects - a lone wooden chair, a solitary stone staircase, and a single window frame - are scattered across the desert, yet they cast no shadows at all, appearing disconnected from the laws of light. The image emphasizes a sense of reality as negative space, where the absent world is overwhelmingly populated and the physical world is sparse. The aesthetic is minimalist and eerie, featuring sharp edges, deep blacks against blinding whites, and a sense of profound, haunting mystery.
• The central figure in the image is a classic, stern profile of a contemplative Sherlock Holmes, depicted from the waist up in a cloak and his famous hat, with a smoking pipe in hand. The image is built on the principle of double exposure, where the lower part of the detective's profile artistically includes a depiction of a Victorian-era London street at night, along which, in the haze and fog of the night, a darkening cab moves away from us. It's an atmospheric and somewhat eerie image, combining the smooth transition of the rising fog and haze of the London street with the smoke from Sherlock Holmes' pipe, depicted against the blurred background of his office. Literary dreams, the cinematic feel of the picture, and the powerful style of the best visualizations of the famous detective’s stories aim to create an incredibly stylish image.
• colossal red torii gate on a jagged black cliff, solitary warrior silhouetted centrally beneath the archway, stone lanterns flanking the path, massive black sphere dominating a vibrant crimson sky, dark rocky foreground terrain, minimalist two-tone palette of pure black and blood red, high-contrast graphic illustration
• an ukiyo-e woodblock print of traffic including a cybertruck on the golden gate bridge.
• Image that is super artsy and Japanese painting stylized. Image looks like it was painted on a piece of bamboo parchment. In the background there is a big red circle (resembling the sun) and black silhouette of Fiji mountains with some black silhouettes of sakura trees. In the foreground there are Japanese stylized wojaks dressed as samurai. They are looking at viewer pogging with their mouths open and they are pointing back at the mountain with outstretched arms. On the side of the image there are three japanese words written verically "最も洗練された", "最も印象深い" and "傑作"
• analog image, film grain, saturated color, high contrast long shadows, artistic framing, almost cinematic feel, This minimalist digital illustration captures a charming animal scene peeking over a solid ledge, shifting the mood to something cute, friendly, and filled with character. The black cat's silhouette remains a dark, soft-edged shadow with alert, pointed ears framing the center-left of the composition. However, its large, round eyes now glow with a vibrant gradient that incorporates deep violet and electric purple tones, narrowing slightly into a pair of thin, vertical black pupils that give the feline a slightly annoyed, grumpy expression. A tiny, luminous magenta nose sits beneath its gaze, adding a whimsical touch to its comical irritation. Sitting right beside the cat is an adorable, fluffy corgi silhouette that brings a burst of friendly warmth to the piece. The corgi's signature large, rounded ears perk up joyfully, and its head is tilted slightly with a soft, bright smile evident in its silhouette. A heavy, stippled noise texture coats both animals, giving their edges a fuzzy, airbrushed quality that feels like a cozy, vintage risograph print or a grainy pastel drawing, perfectly maintaining the signature retro art style while leaning into a playful dynamic. The background frames the duo with a smooth, horizontal gradient sky that evokes a serene, colorful sunset. The top half of the frame is a calm turquoise blue that blends seamlessly down into a radiant, warm coral pink, casting a soft aura behind the contrasting animal shapes. The lower foreground ledge mirrors these colorful tones, shaded in deep purple and violet hues that ground the artwork in a cohesive, neon-infused color palette, turning the scene into a delightful, story-filled interaction between a grumpy cat and a happy pup.
• A highly detailed, professional real-life photograph taken on a Sony A7R IV with a 50mm f/1.8 lens, showing a cozy, cluttered wooden desk. A tiny, adorable chibi dragon, hand-drawn directly onto the surface of a real paper notebook in a rough, smudged charcoal pencil sketch style, has physically come to life. The flat, 2D black-and-white charcoal dragon has crawled partially off the page, and from its open mouth bursts a stream of fire rendered in the same rough, expressive charcoal style - thick smudged lines, sketchy strokes, and messy charcoal dust swirling outward to form the flame. This charcoal-drawn flame billows toward a real, fluffy marshmallow on a toothpick held by a real human hand in the foreground. The warm, intense light from the flame casts dramatic, flickering orange highlights on the charcoal sketch lines and the textured paper, creating a stunning, cross-dimensional depth of field that blurs the background.
• Aerial top-down drone shot capturing the vibrant, dynamic interplay of nature's elements: a narrow, golden-orange sandy beach sharply bordered by dense, dark green coastal vegetation on the left and churning turquoise ocean water with frothy white waves crashing against the shore on the right; the scene is rendered in hyperrealistic photorealism with high detail, sharp focus, and vivid color saturation - emphasizing the stark contrast between the warm earth tones of the sand and the cool, textured blues and greens of the sea, with visible underwater coral or reef structures scattered throughout the shallow waters, under bright, direct daylight with minimal shadows, conveying an intense, immersive, and visually arresting natural landscape devoid of human presence.,
• [Medium and Style] Surreal mixed-media digital collage with a dark cinematic poster aesthetic, vertical square-cropped composition, grainy analog texture, scratched film overlay, high-contrast experimental lighting. The image combines photographic elements with glowing abstract light effects, creating a dreamlike, symbolic, melancholic visual style. [Subject(s) and Clothing] A large glowing human silhouette occupies the left-center of the image, rendered as an overexposed white figure with a vivid electric-blue aura. The silhouette appears faceless and featureless, shaped like a person walking leftward, with a rounded head, torso, arm, loose pants, and bright shoes. The body is blown out into pure white with soft cyan and pale yellow diffusion inside the glow. On the right side, a second human figure stands partly hidden behind a tall vertical blue textured panel or doorway. This figure wears a reddish-brown oversized jacket or coat, loose white pants, and light-colored sneakers. The person has medium-length hair and is shown in profile or partial side view, with the upper body leaning or positioned behind the panel. Their face is partially obscured by shadow and the vertical blue structure. [Pose and Expression] The glowing white figure appears to be walking toward the left, one leg extended forward and one arm hanging downward, but the form is blurred and abstract. The right-side figure stands still behind the vertical panel, with one leg visible and bent slightly, creating a paused, hidden, or transitional posture. The expression is not clearly visible due to shadow, distance, and partial concealment. [Background and Environment] The background is almost entirely black, with a rough, worn texture resembling scratched photographic film or a distressed dark wall. Thin pale scratch marks, dust specks, and vertical streaks are scattered across the image, especially along the right edge and upper area. A rectangular blue light field stretches behind the glowing figure, fading into soft cyan and white near the center. The tall vertical blue panel on the right divides the glowing silhouette from the partially hidden figure, functioning like a doorway, screen, or barrier. [Lighting, Atmosphere and Colors] Low-key lighting dominated by black shadows and a powerful white-blue glow. The glowing figure emits intense white light with saturated cobalt-blue edges, creating a halo that spills across the center of the image. The right figure is dimly lit, with muted reddish-brown, off-white, gray, and beige tones. The palette is mostly black, electric blue, white, cyan, reddish brown, and muted gray. The mood is introspective, lonely, mysterious, and symbolic, suggesting separation, transformation, or leaving behind a former self. [Text and Typography] Small italic lowercase text near the top center reads "goodbye to the person I no longer am". The text is bright cyan-blue, slightly glowing, and set in a clean sans-serif italic style.
• surreal composite photograph, a giant dark skinned man in an oversized charcoal double-breasted suit and deeply pleated wide-leg trousers striding as a towering figure over a miniature San Francisco cityscape, one worn brown suede derby shoe lifted mid-stride looming enormous in the foreground close to the camera, the tiny city spread below - recognizable San Francisco landmarks rendered in miniature including the Transamerica Pyramid, Salesforce Tower, Victorian painted ladies, and steep hilly streets with cable car tracks, tiny yellow cabs and cars dotting the roads below, black beanie, narrow dark sunglasses, woven black crossbody bag, black crow perched on shoulder wings slightly spread, figure looking coolly downward or straight ahead, bright vivid sunny blue sky, sharp photorealistic composite, wide angle lens with strong perspective distortion, cinematic scale and grandeur
• This is a digital magazine cover with a dark, surreal theme. The dominant color is a deep, fiery orange, creating a dramatic and intense background. At the center, a partially obscured female face is visible, with only her eyes and upper nose showing through a black, jagged, and textured pattern that resembles sharp, pointed teeth or spikes. The black pattern is positioned horizontally across the middle of the cover, partially covering the woman's face. Her eyes are wide and expressive, with a slightly anxious or intense look. The overall style is dark, provocative, and thought-provoking, with a strong emphasis on identity and perception.
• A striking, high-contrast graphic poster features the stark black silhouette of a cat's head peeking over a textured horizontal barrier. The composition is defined by its minimalist yet aggressive design, dominated by a raw, textured crimson red background. The cat's fur blends completely into the shadows, with only its pointed ears and piercing gaze defined against the backdrop. Its left eye glows with an intense, vivid yellow hue, while its right eye is cast in a darker, shadowed orange tone, creating a mesmerizing and slightly ominous focal point. The artwork heavily incorporates a gritty, low-fidelity analog texture with deep grain, fine scratches, and subtle print distortions. The red barrier in the foreground features heavy black ink distressing, mimicking a raw woodcut or screen-printed punk zine aesthetic. The clean, sharp geometry of the cat's silhouette clashing with the chaotic, stained textures of the background gives the overall piece a powerful, contemporary graphic design edge.
• A sophisticated concept art illustration of a contemporary architectural masterpiece that blends technical precision with atmospheric artistry. The image combines the clean lines of a blueprint with the expressive textures of a moody digital painting. It depicts a multi-tiered urban sanctuary characterized by sweeping curves and sharp, geometric planes. Fine charcoal lines and delicate ink sketches outline the structural framework, while soft watercolor washes of deep indigo and warm ochre define the shadows and highlights. The building features expansive glass facades that reflect a vibrant, hazy cityscape. Layers of transparency reveal the interior layout, showing floating walkways and lush indoor forests. The composition uses a vertical perspective, looking upward from a bustling plaza, making the structure feel monumental and towering. Golden light spills from the windows, illuminating the surrounding mist and creating a sense of tranquil sanctuary amidst a high-tech metropolis.

System

• Redrob Image

  1.8s

  per image

  Open-weight Z-Image Turbo merge (Apache-2.0, free for commercial use); ~1.8s/image on g7e.2xlarge (NVIDIA RTX PRO 6000 Blackwell Server Edition, 96 GB) after residence, queue excluded; 44-16-30 across 90 blind pairs vs GPT Image 2 (cannot establish parity); stronger skin/light/texture than bare base; strong on photo/portrait/mood; unit cost roughly 1/40th API list at measured utilization, not like-for-like