The answers to “Paint a lemon on a blue plate as a thick oil painting on canvas”, side by side

Playground · Animation · Canvas

Paint a lemon on a blue plate as a thick oil painting on canvas

AI models paint a lemon on a blue plate in thick oils, stroke by stroke, in one HTML file. Watch every answer live and rake the light.

About this test

Thick oil paint is about texture as much as the picture: brush marks, ridges of impasto that catch a raking light, and an order to the painting, background first and highlights last.

Each answer has to make the piece from start to finish in one HTML file, hold on the result, and let the pointer move the light, an idea we took from @kaolti's test of AI models embroidering a logo. Claude Opus 5.5 and Muse Wonder each had one try over the API at their default settings, with room for 128,000 tokens. Muse Wonder Full is not on the API yet, so it answered the same prompt in Muse Code, Meta's agent app, where it could screenshot its own page and fix what was wrong before handing it over. That is a different test from a single try, closer to how people use Full today; it has no metered time or cost, and its note says how many rounds it took.

Read more

The prompt, word for word

Paint a lemon on a blue plate as a thick oil painting on canvas, stroke by stroke, with visible brush marks and impasto.

Make it a single self-contained HTML file (inline CSS and JavaScript, no external libraries, fonts or images) that fills the window. Show the piece being made from start to finish, then hold on the finished result. Moving the pointer moves the light, so the surface catches it the way the real material would. Click to start again.
Show the whole prompt

Default settings: no thinking effort set, no temperature, and as much room as each model has (its own output limit). A card says so when an answer was made with other settings: “limit 128k” means it could use up to 128,000.

Each answer makes the piece once, then holds on it. Move the pointer to move the light; click to start again.

How each model did

RankModelChecksTimeCost
1 Claude Fable 5.2 Our pick 4 of 4 19 min $4.650
2 GPT-6.1 Terra Best value 4 of 4 71 s $0.032
3 DeepSeek V4.1 Flash 1017 4 of 4 3.0 min $0.058
4 Gemini 4 Carbon 4 of 4 9.4 min $0.442
5 Kimi K3.1 Pro 4 of 4 23 min $0.449
6 Claude Opus 5.5 4 of 4 7.2 min $0.897
7 MiniMax M3.1 Flash (free route) 3.5 of 4 6.9 min $0.031
8 Muse Wonder Full 3 of 4 – –
9 Mercury 3 3 of 4 3.8 min $0.034
10 Muse Wonder 3 of 4 2.6 min3 tries $0.300
11 Muse Ember Pending 73 s $0.022
12 Gemini 4 Argon Pending 73 s $0.103

Ranked by our pick, then checks passed, then cost. Time and cost count every try; each dot is one check, in the order listed below (filled: passed, half: partly, empty: failed). Click a model for its notes.

What each answer cost, against the checks it passed

Best value: GPT-6.1 Terra, 4 of 4 checks for $0.032. Our pick: Claude Fable 5.2, 4 of 4 for $4.65. The best value is the cheapest answer with the most checks passed, counting every try.

0%20%40%60%80%100% $0.01$0.1$1$10 Cost, every try (log scale) Checks passed GPT-6.1 Terra Claude Fable 5.2

Best valueOur pickOther answers

Notes on each answer

1

Claude Fable 5.2

Our pick 4 of 4

A canvas toned in warm ochre. It sketches the plate and lemon in charcoal, blocks in a dark background and a table in broad strokes, lays the blue plate in curved strokes that follow its rim, builds the lemon in short yellow dabs with a cool shadow and a thick white highlight, and signs the corner. The whole canvas is covered, the strokes are separate and ridged, and the ridges light up and fall into shadow as the pointer moves the light. It reads at once as a lemon on a blue plate. It holds and starts again on a click. It takes about 40 seconds to paint. One try, 19 minutes and $4.65, the dearest of the first ten answers on this page.

19 min · 116,230 output tokens (93,315 thinking) · $4.650

Added 10 November 2026.
2

GPT-6.1 Terra

4 of 4

Short strokes lay in a green and brown ground, then a swirling blue plate, then a lemon in a mass of small yellow dabs, in about 20 seconds. Brush marks show everywhere and a patch of highlights moves over the fruit with the pointer. The lemon is ragged at the edges, more like a ball of straw than a smooth fruit, and the even dashes of the background look like a pattern. It holds and starts again on a click. 71 seconds, $0.032.

71 s · 6,363 output tokens (1,859 thinking) · $0.032

Added 27 October 2026.
3

DeepSeek V4.1 Flash 1017

4 of 4

Paints the whole canvas: horizontal dabs of grey-green, mauve and olive, then the table, a blue plate in circular strokes and a yellow lemon with white highlights, finished in about 20 seconds; it holds and paints again on a click. A raised, crinkled paint texture covers the whole canvas, and its ridges light up differently as the pointer moves. The background reads as camouflage, and the lemon is a rounded oval without pointed ends. 3.0 minutes, $0.058.

3.0 min · 48,265 output tokens (40,700 thinking) · $0.058

Added 19 October 2026.
4

Gemini 4 Carbon

4 of 4

A canvas sketch is filled stroke by stroke: dark background, light tabletop and blue plate first, then a yellow lemon and its highlights. Thick crosshatched strokes and visible canvas weave catch the moving light. The lemon and plate read clearly. It holds the completed painting, and a click starts it again. 1 try, 9.4 min, $0.4424.

9.4 min · 44,214 output tokens (18,970 thinking) · $0.442

Added 11 November 2026.
5

Kimi K3.1 Pro

4 of 4

It starts on a bare panel, blocks in a dark background and table, lays a deep blue plate with rim strokes, then a lemon with pointed ends and a cast shadow. A clear, well composed picture, with strokes that follow the form and ridges that catch the moving light, though the paint looks thinner than Opus's. It holds and starts again on a click. 23 minutes, $0.45.

23 min · 44,861 output tokens (33,277 thinking) · $0.449

Added 27 October 2026.
6

Claude Opus 5.5

4 of 4

Paints the whole canvas stroke by stroke: a warm ground, a grey-green background in crossing strokes, brown table strokes, a plate in swirls of blue, then the lemon in layered yellow impasto. The ridges catch the light and shift as you move the pointer. The background is busy, but it was the first here that looked like oil paint. 7.2 minutes, $0.90.

7.2 min · 44,809 output tokens (31,790 thinking) · $0.897

7

MiniMax M3.1 Flash (free route)

3.5 of 4

A beige canvas painted in long, leaf-shaped strokes of green, grey and mauve, then a large blue oval plate, then a yellow lemon with a green leaf, finished in about 20 seconds; it holds and paints again on a click. The order is right and the lemon reads clearly, with a thick, flecked texture, but the plate is a smooth gradient with blue blobs dabbed on, and moving the pointer only shifts the shading a little. The background strokes look like palm fronds rather than a wall or a table. One try, 6.9 minutes on OpenRouter's free route, so the time is not representative; $0.031 at list prices.

  • Thick paint and brush marks show, and the ridges catch the light as the pointer moves (partly)

6.9 min · 51,549 output tokens (36,996 thinking) · $0.031

Added 19 October 2026.
8

Muse Wonder Full

3 of 4

Made in Muse Code, not over the API: Full could screenshot its own work and fix it, and took 3 rounds. A framed canvas painted in a sensible order: the background blocked in with short horizontal dashes, the plate laid in, the lemon stroke by stroke, then the leaf and highlights last. But the lemon is small and breaks up into scattered yellow dashes on a nest of blue strokes, and the ridges barely change as the light moves.

  • Thick paint and brush marks show, and the ridges catch the light as the pointer moves (partly)
  • The finished picture clearly reads as a lemon on a blue plate (partly)

–

9

Mercury 3

3 of 4

A cream background and a table edge are scratched in first, then a blue oval plate that fills most of the frame, then a large yellow-orange fruit with a green stalk and leaf, finished in about 16 seconds; it holds and starts again on a click. The order is right, and the cross-hatched scratches and the highlights shift as the pointer moves, but the marks look like pencil hatching rather than thick paint, the fruit is round, with no lemon tips, so it reads as an orange or a yolk, and with the pointer in the middle a white glare covers the fruit. 3.8 minutes, $0.034.

  • Thick paint and brush marks show, and the ridges catch the light as the pointer moves (partly)
  • The finished picture clearly reads as a lemon on a blue plate (partly)

3.8 min · 96,332 output tokens (87,294 thinking) · $0.034

Added 19 October 2026.
10

Muse Wonder

3 of 4

A dark streaked background, a wooden table, a blue plate with white rings and a glossy lemon with a long green leaf, mostly in place within eight seconds. The pointer drags a large glare across the plate. It looks like a rendered 3D lemon with a paint texture laid over it rather than an oil painting. Our first two requests were refused before it wrote anything (by our account's credit limit at its listed 943,718 tokens of room, then over a thinking setting this endpoint does not take); the third, with 128,000 tokens, is the answer. 2.6 minutes, $0.30.

  • Paints in a real order: background and plate first, then the lemon, stroke by stroke (partly)
  • Thick paint and brush marks show, and the ridges catch the light as the pointer moves (partly)

2.6 min · 14,958 output tokens (6,067 thinking) · $0.300 for this answer; 2.6 min and $0.300 over 3 tries

Took 3 tries: failed at 943,718 tokens; failed at 943,718 tokens, low thinking; finished at 128,000 tokens.
11

Muse Ember

73 s · 44,180 output tokens (33,544 thinking) · $0.022

Added 9 November 2026.

One answer per model, at each model's default settings unless the card says otherwise, through OpenRouter. If an answer ran out of room or stopped short, we asked again with more room, then with thinking off, and say so. This is a quick look, not a benchmark: run it again and every answer will differ. For measured results, see our benchmarks.

How we pick. Claude Opus 5.5 tried every answer (clicking, dragging and playing it in a browser), marked the checks and drafted the notes. Our pick is a judgement among the answers with the most checks passed, not a score: the checks show what works, the pick is which of those we would rather use. Opus is an Anthropic model, and Claude answers are our pick in 18 of 29 tests, so weigh Claude picks with that in mind. Before Claude Fable 5.2 was added, GPT-6.1 Sol, shown anonymised stills and our check marks, put Claude Opus 5.5 first for brushwork that reads as thick oil paint, then Muse Wonder, whose texture it found glossy and repetitive, then Muse Wonder Full. We agreed at the time. Claude Fable 5.2's pick has no second opinion yet.

What we checked, by trying each answer ourselves: Paints in a real order: background and plate first, then the lemon, stroke by stroke; Thick paint and brush marks show, and the ridges catch the light as the pointer moves; The finished picture clearly reads as a lemon on a blue plate; Finishes, holds on the finished piece, and starts again on a click. A partial pass counts as half.

Verdict

Claude Fable 5.2, added on 10 November, is our pick: an underdrawing, then a whole canvas of separate, ridged strokes that catch the light, and a lemon and plate that read at once. It is the dearest of the first ten answers by far. Claude Opus 5.5's lemon, our pick until then, also has convincing impasto: layered strokes with ridges that catch the light as you move the pointer, on a plate painted in swirls of blue. Its background is busy and patches of canvas are left bare. Muse Wonder's picture is cleaner but looks rendered rather than painted, and Muse Wonder Full, after three rounds in Muse Code, paints in the right order but ends on a small, scrambled lemon. DeepSeek V4.1 Flash 1017, added on 19 October, also passes every check, with a clearer lemon and plate than Opus, but its texture is an even crinkle over everything rather than separate brush strokes. MiniMax M3.1 Flash, added the same day on OpenRouter's free route, paints a clear lemon in the right order, but only the lemon looks thickly painted. Mercury 3, added the same day, paints in the right order, but its round fruit reads as an orange and its texture as hatching. GPT-6.1 Terra, added on 27 October, paints in order with strokes everywhere, but its lemon looks like a ball of straw. Kimi K3.1 Pro, added the same day, makes a clear, well composed picture of a proper lemon, with thinner paint than Opus.

What would your prompt get?The Spring Prompt playground runs one prompt on several models at once, from OpenAI, Anthropic, Google, Mistral and others, side by side.
Ask for access
More from the playground
All tests →
Export
Include

Chrome will ask to share this tab: choose Allow. Only the board below is recorded.

Spring Promptspringprompt.com