Lucy’s Artistly Field NotesSubmit a new study ↗I test Artistly’s image tools and styles the same way, over and over, to see where they follow a prompt, where they drift, and which defaults quietly shape the result. This is that notebook.
Lucy’s research desk ✦The Seamless Patterns v2 tool generates 2 images per prompt submission. The left slot is the cinematic one: strong perspective and color grading, and it recognizes named art styles like Jack Kirby's, but it garbles text and sometimes swaps in its own details, like mugs for teacups. The right slot is the literal one: it renders requested text perfectly and repeats itself closely run after run, but it recognizes named styles less well and tends to add a caption or signature nobody asked for. My best guess, at moderate confidence, is that the right slot is Bria FIBO and the left is a broad-web model, possibly FLUX or SDXL-based. Artistly hasn't confirmed either.
Captures the spirit of a prompt: dramatic perspective, heavy color grading, and it recognizes named artist styles. It struggles with text and sometimes swaps in its own details, like mugs for teacups.
Follows the wording of a prompt: perfect requested text, exact details, and nearly identical results run after run. It recognizes named styles less well and tends to turn a name into visible text.
Same prompt and aspect ratio, run 3–4 times per test. I scored each slot by what it actually produced, not by which image looked nicer.
This one started it all. Image models are notoriously bad at self-reference, but naming the art term (‘Droste effect’) and spelling out the nested content worked for both slots. For comparison, Artistly's AI Image Designer v6 dropped the concept entirely and drew a plain landscape. 1 run.
Recursion achieved, with a looser, more painterly watercolor feel.
Recursion achieved, cleaner and more consistent, with the nested figure clearly the same woman.
Pencil sketch illustration of a woman painting at an easel in an open landscape. On her canvas, clearly visible, is a small watercolor painting of the same scene: the same woman, at the same easel, in the same landscape, painting. Droste effect, recursive image, picture within a picture, graphite and soft gray watercolor washes, white background.
A vintage hand-painted hardware-store sign asked for an exact, oddly specific line of text. This was the clearest separation between slots. 3 runs.
Garbled lettering in every run, including ‘SCMALL’, ‘J9B’, and the date turning into 1924.
Correct requested text in all three runs, down to the decorative dashes.
A vintage hand-painted wooden shop sign hanging outside a small hardware store. The sign reads exactly: "EDJ'S GEAR & GIZMO EMPORIUM — EST. 1972 — NO JOB TOO SMALL". Cream lettering on a deep green background, weathered paint, brass hanging brackets.
The burnt-orange mug prompt tested perspective, color, and a light source from camera-right. 4 runs.
Stronger low-angle, wide-angle perspective; shadows sometimes missed the requested direction.
Shadow fell left in all four runs; the lens effect was milder.
A ceramic coffee mug on a wooden workbench. Camera: extreme low angle at table height, 24mm wide-angle lens, strong perspective distortion. Lighting: single hard light source from camera-right, casting a long sharp shadow to the left. Background: out of focus. Color palette limited to burnt orange, charcoal gray, and cream.
A Jack Kirby-inspired robot scene exposed the right slot's tendency to draw the style reference as lettering. 4 runs.
Stronger Kirby-like action, armored figures, energy and dot bursts.
Generic comic look; added unrequested speech bubbles and text.
A robot walking a dog through a city park, drawn in the style of Jack Kirby's 1970s Marvel comics — Kirby crackle energy dots, blocky dynamic anatomy, dramatic foreshortening.
A surprisingly strong result for both slots: five differently colored cups in order. 8 images.
Correct count and positions, but teacups often became mugs.
Correct count, order, saucer placement, and mint detail with more textbook teacups.
Exactly five teacups in a straight row on a white shelf. From left to right: red, red, blue, yellow, red. Only the blue cup has a saucer. A single sprig of mint sits in the yellow cup. Plain gray wall behind.
The atmospheric details competed with the requested Norman Rockwell look. 4 runs.
Warm glossy glow overwhelmed the named style.
Closer to period illustration; added border, signature, caption and headline.
A cozy reading nook inspired by Norman Rockwell's Saturday Evening Post covers — a grandfather reading to a young girl by a fireplace, warm lamplight, a sleeping dog on the rug.
Replacing cozy atmosphere with explicit style traits brought both closer to the target. 4 runs.
Clearer humor and more expressive faces; missed the braided round rug.
Kept listed details, including the rug; grandpa looked awake and a signature appeared.
A humorous everyday moment painted in the style of Norman Rockwell: an elderly grandfather in a worn armchair has fallen asleep mid-bedtime story, mouth open, reading glasses slipping down his nose, the open storybook sliding off his lap. His wide-awake granddaughter, in pajamas and fuzzy slippers, has taken over — standing on the rug, holding the book up with great seriousness and reading aloud to the family beagle, who sits upright listening intently with his head tilted. Exaggerated, expressive faces full of personality. Only the armchair, a small braided rug, and a side table with a teacup — everything else is a plain off-white background. Crisp, realistic oil illustration, precise detail, muted mid-century palette, soft even lighting, warm and gently funny.
Both slots followed the structured prompt well. This does not establish that either receives JSON directly. 4 runs.
Fixed mug color and shadow direction while retaining dramatic perspective.
Kept detail accuracy, with a milder lens effect.
{"short_description": "A ceramic coffee mug on a wooden workbench, photographed from table height with dramatic wide-angle perspective.", "objects": [{"description": "A plain ceramic coffee mug with a handle", "location": "center", "relative_size": "medium within frame", "shape_and_color": "Cylindrical, burnt orange with a cream rim", "orientation": "Handle facing right"}], "background_setting": "An out-of-focus workshop behind the workbench", "lighting": {"conditions": "Single hard light source", "direction": "From camera-right", "shadows": "One long, sharp shadow falling to the left of the mug"}, "aesthetics": {"composition": "Mug centered, workbench surface stretching toward the viewer", "color_scheme": "Limited to burnt orange, charcoal gray, and cream", "mood_atmosphere": "Quiet, dramatic"}, "photographic_characteristics": {"depth_of_field": "Shallow", "focus": "Sharp on the mug", "camera_angle": "Extreme low angle at table height", "lens_focal_length": "24mm wide-angle"}, "style_medium": "photograph", "text_render": [], "artistic_style": "Realistic"}
Adapted from my write-up, Two Models, One Button: Fingerprinting Artistly's Seamless Patterns v2 (Sep 24, 2026). The original draft marks some image pairs and aspect ratios as pending; those examples are described here without inventing images or ratios that weren't captured.
Pick a prompt close to the work you actually make. Run it 3–4 times with the same words and ratio, and record what each image gets right and wrong.
Change one thing at a time. Test text, counting, lighting, style, or another single behavior.
Hold the setup steady. Keep wording, tool, and aspect ratio identical across runs.
Count the failures. Log specific misses by left and right slot; avoid judging only by which looks nicer.
Share the raw details. Include the exact prompt, number of runs, ratio, and image link if you have one.
A set of 6 stickers featuring the same cute cartoon [ANIMAL] with a [DISTINCT ACCESSORY], shown [SIX ACTIVITIES]. Thick white sticker border around each one, soft cartoon style, clean outlines, plain neutral background. The same character in every sticker.
Run multiple repetitions before drawing a conclusion. Tools and settings can change after publication.
Reviewed observations from people who ran their own version of this test, with their prompt and evidence.
Test submission only, not a real finding. Please reject or delete this once you've confirmed it appears correctly in your review queue and, if approved, in the community log.
This is a placeholder test entry submitted to verify the community findings pipeline (form, database, review queue, and public display) before inviting real testers. No actual Seamless Patterns v2 runs were performed for this entry.
Your observation will be linked to Case Study: Seamless Patterns v2 after I review it. If you have a different question in mind, submit a new case study instead.
These are informal community tests, not a benchmark. 3–4 runs per test is enough to spot strong patterns, not subtle ones, and Artistly can swap models, change settings, or reorder the slots without notice. I assumed the same model always appears on the same side; every test so far supports that, but it's worth rechecking now and then. A slot's behavior can suggest a model identity, but it can't prove one.
Writing the mug prompt as structured JSON made the two slots agree far more than plain English did, on color, orientation, and lighting. Whether Artistly passes that JSON straight to either model, or runs it through a translator first, I still don't know. My best guess is that FIBO's pipeline converts plain prompts into structured fields before generating, and that a dedicated text-render field is what causes the right slot's unrequested captions and signatures when a style reference implies text, but that's speculation, not something I've observed directly.
FIBO documentation ↗SDXL paper ↗FLUX.2 repository ↗