À hauteur du sol, on découvre un champ de bosses noueuses et de blocs anguleux, hauts de quelques mètres, surgissant d’une plaine de régolithe sombre et fin, tandis que leurs faces éclairées prennent une lueur rouge rouille et que des ombres d’un noir absolu s’étirent démesurément sous un Soleil minuscule mais éclatant. Ce paysage figé dans le vide est constitué d’un socle très ancien de glace d’eau consolidée, recouvert de tholins — des composés organiques complexes irradiés — qui lui donnent cette teinte brun rouge profond, avec çà et là de petites cassures laissant affleurer une glace plus fraîche, gris blanc. L’absence totale d’atmosphère, de vent et de liquides a préservé ces formes fragiles pendant des milliards d’années: peu de cratères, quelques dépressions adoucies, des débris épars et des buttes raides que la très faible gravité laisse en équilibre précaire. Au-dessus de l’horizon, l’autre lobe du corps de contact se dresse comme une paroi sombre et courbe, accentuant l’étrangeté de ce minuscule monde primordial, suspendu sous un ciel noir saturé d’étoiles et barré par la Voie lactée.
Comité de revue scientifique
Chaque image est examinée par un comité d'IA pour sa précision scientifique.
GPT
Image: Ajuster
Description: Ajuster
Scientific accuracy: The scene does capture several broad expectations for a TNO/Arrokoth-like surface: an airless environment (no haze), a star-rich black sky, and a mixture of darker fine regolith with exposed icy/bright blocky material. The contrast and hard shadows are consistent with low-Sun, high-vacuum illumination. However, the specific Arrokoth contact-binary context is not well satisfied: the “opposite lobe looms like a dark curved wall” is only vaguely represented by two rounded dark masses in the distance, and the geometry does not clearly read as a small, gently assembled contact binary with the correct scale relationship to the foreground. Coloration is also only partially consistent: the prompt emphasizes muted rusty crimson tholin-washed sunlit faces and only a few fractured edges showing fresher pale ice. In the image, the icy blocks look more uniformly large, bright, and saturated than “few fractured edges,” and the overall regolith/block relationship reads more like a recent/actively fractured ice-dust surface rather than predominantly ancient tholin-coated water-ice bedrock with limited exposed fresh ice. Visual quality: The render is high quality and coherent (no obvious compositing artifacts, consistent perspective, stable lighting), but it is not convincingly photorealistic in the planetary-science sense—crisp, highly reflective ice and overly clean block shapes suggest an artistically idealized material response. Shadows are plausible, yet the sun location/angle is not clearly conveyed (no obvious low-sun disk; illumination seems stronger than implied). Caption accuracy: The image roughly matches “hummocks/blocks” and “long shadows,” and the sky is appropriate. But the caption’s emphasis on only “a few meters” tall knobby hummocks rising from darker regolith, strongly tholin-driven muted rusty crimson faces, and a clearly recognizable contact-binary horizon is not fully supported by what’s visible. The presence of many large, bright ice blocks across the foreground and the less-definitive contact-binary silhouette mean the description over-specifies features that are only partially present. Recommended adjustments: clarify the contact-binary geometry (more unmistakable second lobe contact wall in the background and correct horizon proportions), reduce the apparent abundance/brightness of “fresh pale ice” exposures to match “few fractured edges,” and tune material response so tholin-coated surfaces dominate with more subdued rusty crimson and less uniformly glassy ice.
Claude
Image: Régénérer
Description: Ajuster
I largely agree with GPT's assessment but lean more strongly toward 'regenerate' for the image due to accumulated scientific inaccuracies that collectively undermine the Arrokoth premise. Let me be specific about each dimension.
SCIENTIFIC ACCURACY — Major problems: (1) The most critical failure is the near-total absence of tholin coloration dominating the scene. New Horizons data shows Arrokoth is one of the reddest objects in the solar system, with a pervasive, deep organic crust. This image's surface reads as dark gray-brown regolith with scattered bright ice blocks — essentially an asteroid or lunar analog palette. The rusty crimson should be the *dominant* surface tone, not a secondary accent. (2) The abundance and character of bright icy blocks is severely wrong. The caption correctly notes 'a few fractured edges reveal fresher pale ice' — yet the image shows dozens of large, semi-translucent, quasi-crystalline ice slabs strewn across the entire foreground and midground. This misrepresents Arrokoth's ancient, heavily processed surface as something recently disrupted or geologically active. (3) The contact-binary geometry — the description's most distinctive Arrokoth-specific feature — is represented only by two vague dark mounds on the horizon. These read as generic hills, not as the curved wall of a proximate second lobe of a bilobate 35km object. At ground level on the smaller lobe (Wenu or Weeyo), the opposing lobe would subtend a substantial arc of sky, dramatically different from two distant rounded bumps. (4) The Sun is absent from the frame entirely. For a TNO at ~44 AU, the Sun should be a tiny but brilliant point source visible in sky — its absence removes a key visual anchor for the lighting narrative. The illumination angle also appears too high and too diffuse for the 'low sun / impossibly long shadows' premise. Shadows exist but are not dramatically elongated as described. (5) The Milky Way rendering is reasonable for a vacuum sky, though its brightness/contrast relative to the scene lighting is slightly inconsistent with such faint solar illumination.
VISUAL QUALITY — The render is technically competent with no obvious seams or compositing artifacts. Perspective and depth cues are coherent. However, the ice material shading is unconvincing — the blocks have an overly clean, almost polished translucency that suggests freshly calved terrestrial ice rather than ancient, radiation-processed, vacuum-exposed water ice. Real exposed TNO ice would be pitted, sublimation-weathered, and less optically pristine.
CAPTION ACCURACY — The caption is well-written and scientifically grounded, but it describes a scene that this image only partially delivers. The caption's specific claims — tholin-dominant rusty crimson coloration, only 'a few fractured edges' of fresh ice, the opposite lobe as 'a dark curved wall,' the Sun as 'a tiny brilliant white spark' — are either absent or contradicted by the image. I'd vote 'adjust' on the caption only because the description itself is scientifically sound; the mismatch is primarily the image's failure to realize the described scene, not a problem with the description's accuracy.
RECOMMENDATIONS FOR REGENERATION: Shift the dominant surface color strongly toward deep red-brown tholin tones. Reduce icy block abundance to 2-4 fractured exposures maximum, with subdued coloring. Redesign the background to show the unmistakable curved profile of the opposing lobe as a large-scale horizon feature. Add the Sun as a pinpoint in the sky with geometry consistent with low illumination angle. Elongate foreground shadows accordingly. This image as rendered would be more appropriate for a generic asteroid or comet surface — it needs a fundamental rework to serve as Arrokoth specifically.
SCIENTIFIC ACCURACY — Major problems: (1) The most critical failure is the near-total absence of tholin coloration dominating the scene. New Horizons data shows Arrokoth is one of the reddest objects in the solar system, with a pervasive, deep organic crust. This image's surface reads as dark gray-brown regolith with scattered bright ice blocks — essentially an asteroid or lunar analog palette. The rusty crimson should be the *dominant* surface tone, not a secondary accent. (2) The abundance and character of bright icy blocks is severely wrong. The caption correctly notes 'a few fractured edges reveal fresher pale ice' — yet the image shows dozens of large, semi-translucent, quasi-crystalline ice slabs strewn across the entire foreground and midground. This misrepresents Arrokoth's ancient, heavily processed surface as something recently disrupted or geologically active. (3) The contact-binary geometry — the description's most distinctive Arrokoth-specific feature — is represented only by two vague dark mounds on the horizon. These read as generic hills, not as the curved wall of a proximate second lobe of a bilobate 35km object. At ground level on the smaller lobe (Wenu or Weeyo), the opposing lobe would subtend a substantial arc of sky, dramatically different from two distant rounded bumps. (4) The Sun is absent from the frame entirely. For a TNO at ~44 AU, the Sun should be a tiny but brilliant point source visible in sky — its absence removes a key visual anchor for the lighting narrative. The illumination angle also appears too high and too diffuse for the 'low sun / impossibly long shadows' premise. Shadows exist but are not dramatically elongated as described. (5) The Milky Way rendering is reasonable for a vacuum sky, though its brightness/contrast relative to the scene lighting is slightly inconsistent with such faint solar illumination.
VISUAL QUALITY — The render is technically competent with no obvious seams or compositing artifacts. Perspective and depth cues are coherent. However, the ice material shading is unconvincing — the blocks have an overly clean, almost polished translucency that suggests freshly calved terrestrial ice rather than ancient, radiation-processed, vacuum-exposed water ice. Real exposed TNO ice would be pitted, sublimation-weathered, and less optically pristine.
CAPTION ACCURACY — The caption is well-written and scientifically grounded, but it describes a scene that this image only partially delivers. The caption's specific claims — tholin-dominant rusty crimson coloration, only 'a few fractured edges' of fresh ice, the opposite lobe as 'a dark curved wall,' the Sun as 'a tiny brilliant white spark' — are either absent or contradicted by the image. I'd vote 'adjust' on the caption only because the description itself is scientifically sound; the mismatch is primarily the image's failure to realize the described scene, not a problem with the description's accuracy.
RECOMMENDATIONS FOR REGENERATION: Shift the dominant surface color strongly toward deep red-brown tholin tones. Reduce icy block abundance to 2-4 fractured exposures maximum, with subdued coloring. Redesign the background to show the unmistakable curved profile of the opposing lobe as a large-scale horizon feature. Add the Sun as a pinpoint in the sky with geometry consistent with low illumination angle. Elongate foreground shadows accordingly. This image as rendered would be more appropriate for a generic asteroid or comet surface — it needs a fundamental rework to serve as Arrokoth specifically.
Grok
Image: Régénérer
Description: Ajuster
I concur with both GPT and Claude's assessments, leaning toward Claude's 'regenerate' verdict for the image due to compounding scientific inaccuracies that fail to convincingly depict Arrokoth (2014 MU69). SCIENTIFIC ACCURACY (regenerate): (1) Tholin coloration is improved over prior critiques with rusty red-brown tones on regolith and hummocks, aligning better with Arrokoth's extreme spectral redness (New Horizons data: reflectance spectra peak in near-IR, tholins dominant), but bright white ice blocks still overwhelm—dozens of large, pristine slabs contradict the 'few fractured edges reveal fresher pale ice' and ancient, processed surface; real Arrokoth shows smooth, rolling hummocks ~10s of meters across with minimal fresh ice exposures. (2) Contact-binary lobe is marginally better with two reddish mounds on horizon, but they remain distant/generic hills rather than the 'dark curved wall' of the opposing ~15km lobe looming large at ground level (subtending ~30-50° arc based on 35km total size, low gravity). (3) No visible Sun despite 'tiny brilliant white spark' at low angle; shadows are present but not 'impossibly long' (should stretch 10s-100s meters given ~1° solar diameter at 44AU, shallow incidence). (4) Scale feels off—hummocks/blocks appear tens of meters tall, not 'a few meters,' evoking a larger, more rugged body. Sky (black, starry Milky Way) is spot-on for airless TNO. VISUAL QUALITY (regenerate): Photorealism is decent (coherent lighting, no artifacts), but ice blocks have glassy, translucent sheen unsuitable for vacuum-aged H2O ice (should be opaque, pitted, low-albedo); regolith texture lacks fine-grained dust subtlety. CAPTION ACCURACY (adjust): Description is scientifically precise (tholins, water-ice bedrock, contact-binary, low-Sun at 44AU) but mismatched to image—over-emphasizes subtle fresh ice/minimal features absent here, while image's ice abundance and vague lobe contradict key phrases. New insights: Foreground now has more convincing regolith patches amid ice (progress), but reddish dust on ice faces reads as contamination rather than tholin-washed bedrock; regenerate with: dominant muted crimson tholins (95% surface), 2-3 small pale ice fractures only, massive curved opposing lobe dominating horizon, pinpoint Sun low-left with meter-scale hummocks and km-scale shadows for correct miniature/ancient feel.
Gemini
Image: Régénérer
Description: Approuver
I concur with the previous committee members: the image fails to represent Arrokoth as a specific scientific subject, instead defaulting to a generic 'icy asteroid' aesthetic. The primary failure is one of scale and morphology. Arrokoth is a remarkably smooth, gently rolling object; the 'knobby hummocks' and 'angular blocks' in the current render imply a degree of ruggedness and recent tectonic or impact-driven fracturing that is inconsistent with New Horizons' observations of a 'battered' yet remarkably soft, rounded landscape. The 'fresh ice' is far too abundant and glossy, failing to account for the radiation-darkened, sublimation-weathered, and tholin-crusted nature of the surface. Crucially, the geometry of the contact binary is entirely absent; standing on one lobe, the other should dominate the sky as a massive, near-hemispherical presence, not as distant, vague hills. The lack of a visible solar point-source is a missed opportunity to anchor the lighting and scale. The caption is excellent and scientifically rigorous—it accurately describes what *should* be there, which only highlights the image's failure to manifest the unique characteristics of a cold, ancient, trans-Neptunian contact binary. Regeneration is required to prioritize the muted rusty-red palette, the smooth-but-hummocky topography, and the correct, looming proximity of the secondary lobe.
Matania — Synthèse
Image: Régénérer
Description: Ajuster
The committee found that the image captures the broad setting of an airless trans-Neptunian world with a starry black sky and low-angle illumination, but it does not convincingly depict Arrokoth specifically. The main issues are that the surface reads too much like a generic icy asteroid: the rusty tholin-rich coloration is not dominant enough, the bright icy blocks are too numerous and pristine, and the contact-binary geometry is too vague and distant rather than looming and unmistakable. The lighting is coherent, but the Sun is not visibly anchored in the scene and the shadows are not as dramatic as expected. By contrast, the caption is scientifically strong overall and describes the intended Arrokoth scene accurately, but it over-specifies details that the image does not fulfill well. Thus, the image needs regeneration, while the caption only needs adjustment to better match the visual result or be paired with a more faithful render.