Cyberpunk LLMFit 模型仪表盘
生成一张深色科幻风格的仪表盘信息图，展示本地硬件及六种推荐的 LLM 模型配置，适用于机器学习控制面板。
gpt-image-2 · App / 网页设计 · 信息图 / 教育视觉图 · 图表 · 极简主义 · 赛博朋克 / 科幻
中文提示词
目标：创建一个深色未来主义风格的仪表盘信息图，用于本地 AI 模型推荐，标题为 {argument name="headline text" default="LLMFIT RECOMMENDATIONS"}，中央大面板显示 {argument name="main title" default="LEGION MODEL LOADOUT"}。设计应呈现为按需生成的本地机器赛博朋克硬件分析 UI，而非营销海报。 画布：21:9 宽屏横向图像，约 1200x560，黑色与深青色背景，带有微妙的辉光、纤细的青色网格线、淡淡的电路轨迹以及带边框的应用程序窗口。在窗口右上角添加一个小的关闭按钮图标。使用锐利的科幻字体、压缩的大写标题、霓虹青柠色点缀、青色轮廓以及细小的琥珀色注释文本。 顶部标题：左上角显示小标签“LEGION / MODEL INTELLIGENCE”，位于标题上方。在标题下方，显示硬件摘要文本：{argument name="hardware summary" default="NVIDIA GeForce RTX 5090 · 31.84 GB VRAM · 125.18 GB RAM"}。 主要布局：将仪表盘分为两个主要列。左列占据约 70% 的宽度，包含主要的配置卡片。右列占据约 30% 的宽度，包含紧凑的验证数据列表。 左侧主卡片：创建一个带有霓虹边框的大面板，以青柠绿和白色显示大标题“LEGION MODEL LOADOUT”。在其下方，单行显示 4 个硬件能力徽章：1) NVIDIA GeForce RTX 5090，31.8 GB VRAM，配 GPU 风扇图标；2) Intel(R) Core(TM) Ultra 9 285K，配 CPU 芯片图标；3) 125.2 GB 系统 RAM，配内存条图标；4) CUDA，配圆形 CUDA 图标。GPU 徽章使用青柠色，其他使用青色。 配置表格：在徽章下方，显示 6 行排名推荐，带有圆角框内的青柠色大行号和纤细的青色分隔线。每行应包含模型名称、量化方式、运行时、内存和预估速度。6 行内容如下：1) "shawnw3j/Huihui-Qwen3.6-27B-abliterated-AWQ-MTP"，量化 "AWQ-4bit"，运行时 "vLLM"，内存 "14.7 GB"，速度 "80.9 estimated tok/s"；2) "Vortex5/G4-Starry-Ocean-12B"，量化 "Q8_0"，运行时 "llama.cpp"，内存 "16 GB"，速度 "82.8 estimated tok/s"；3) "shawnw3j/Qwen3.6-27B-AWQ-MTP"，量化 "AWQ-4bit"，运行时 "vLLM"，内存 "14.7 GB"，速度 "80.9 estimated tok/s"；4) "Minachist/Qwen3.6-27B-INT8-Autoround-V2"，量化 "AutoRound-4bit"，运行时 "vLLM"，内存 "16.6 GB"，速度 "80.9 estimated tok/s"；5) "exnivo/Qwen3.8-20B-Minitron"，量化 "Q8_0"，运行时 "llama.cpp"，内存 "22.6 GB"，速度 "49.9 estimated tok/s"；6) "Lorbus/Qwen3.6-27B-int4-AutoRound"，量化 "AutoRound-4bit"，运行时 "vLLM"，内存 "16.6 GB"，速度 "80.9 estimated tok/s"。在指标列中添加芯片、终端/运行时、内存和速度计的小图标。 主卡片页脚：居中显示青色文本：“ESTIMATED BY LLMFIT · VERIFY WITH A LOCAL BENCHMARK。”并在其周围添加角括号和纤细的装饰性电路段。 右侧边栏：标题“VERIFIED LLMFIT DATA”位于左侧，小号琥珀色文本“ESTIMATES, NOT BENCHMARKS”位于右侧。显示 6 行与上述推荐对应的紧凑验证数据，编号为 01 至 06（青柠色）。每行显示缩短的模型名称、包含量化/运行时/内存的第二行小字，以及右对齐的大号分数：80.9、82.8、80.9、80.9、49.9、80.9。底部添加一条琥珀色小注：“llmfit recommendations are estimates from detected hardware, not measured benchmarks.” 底部窗口栏：左下角添加微小的时间戳文本“GENERATED 8/17/2026, 7:35:32 PM”。右下角添加一个小的矩形霓虹绿按钮，标签为 {argument name="button label" default="Refresh scan"}，并配有刷新图标。 视觉约束：保持所有文本为英文，清晰易读，除指定的 6 行推荐外不添加额外行，除指定的 4 个硬件徽章外不添加额外徽章。使用深色透明玻璃 UI 风格，带有微妙的辉光效果，画面中不包含人物，除文本硬件/模型标签外不包含任何 Logo，且无水印。
原始提示词
Goal: Create a dark futuristic dashboard infographic for {argument name="dashboard title" default="LLMFIT RECOMMENDATIONS"}, showing local AI model recommendations for a workstation. Canvas: Wide 21:9 desktop-panel image, black and deep teal background with subtle glow, thin neon cyan border, faint scanline/grid texture, compact technical UI styling. Layout: Top header bar with small label “LEGION / MODEL INTELLIGENCE”, large title “LLMFIT RECOMMENDATIONS”, subtitle “NVIDIA GeForce RTX 5090 · 31.84 GB VRAM · 125.18 GB RAM”, and a small outlined close-box icon at the top right. Main content is split into two panels: a large left recommendation module occupying about two thirds of the width, and a narrower right verification panel. Left panel: Add a large cyberpunk card titled “LEGION MODEL LOADOUT”, with “LEGION” in neon lime and the rest in white blocky techno typography. Under the title, show exactly 4 hardware/status tiles with icons: 1) NVIDIA GeForce RTX 5090, 31.8 GB VRAM with a GPU fan icon, 2) Intel(R) Core(TM) Ultra 9 285K with a CPU chip icon, 3) 125.2 GB system RAM with a memory module icon, 4) CUDA with a circular CUDA emblem. Beneath the tiles, show exactly 6 ranked model rows, each with a large lime outlined rank number, model name, quantization, runtime, RAM amount, and estimated tok/s speed. The 6 rows are: 1) shawnw3j/Huihui-Qwen3.6-27B-abliterated-AWQ-MTP — AWQ-4bit — vLLM — 14.7 GB — 80.9 estimated tok/s; 2) Vortex5/G4-Starry-Ocean-12B — Q8_0 — llama.cpp — 16 GB — 82.8 estimated tok/s; 3) shawnw3j/Qwen3.6-27B-AWQ-MTP — AWQ-4bit — vLLM — 14.7 GB — 80.9 estimated tok/s; 4) Minachist/Qwen3.6-27B-INT8-Autoround-V2 — AutoRound-4bit — vLLM — 16.6 GB — 80.9 estimated tok/s; 5) exnivo/Qwen3.8-20B-Minitron — Q8_0 — llama.cpp — 22.6 GB — 49.9 estimated tok/s; 6) Lorbus/Qwen3.6-27B-int4-AutoRound — AutoRound-4bit — vLLM — 16.6 GB — 80.9 estimated tok/s. Use small cyan icons for chip/runtime/RAM/speed columns. Footer inside left card: Center a slim neon divider with the text “ESTIMATED BY LLMFIT · VERIFY WITH A LOCAL BENCHMARK.” Right panel: Title it “VERIFIED LLMFIT DATA” with a small note “ESTIMATES, NOT BENCHMARKS”. Show exactly 6 compact verification entries matching the same 6 models, numbered 01 through 06 in lime, each with smaller gray metadata text and a bright cyan score on the far right: 80.9, 82.8, 80.9, 80.9, 49.9, 80.9. Add a small orange warning note at the bottom: “llmfit recommendations are estimates from detected hardware, not measured benchmarks.” Bottom app chrome: Add a tiny timestamp line at bottom left, “GENERATED 8/17/2026, 7:53:32 PM”, and a small green outlined button at bottom right labeled {argument name="button label" default="Refresh scan"} with a refresh icon. Visual style: High-contrast sci-fi terminal UI, angular panel corners, thin glowing cyan circuit traces, lime accents, white condensed techno font, dense but readable technical typography, subtle green monitor glow. Keep the image crisp like a generated dashboard screenshot, not a poster. Constraints: Use exactly 6 model recommendation rows, exactly 4 hardware tiles, and exactly 6 verified-data entries. Do not add people, photos, logos beyond simple hardware-style icons, or extra sections. Keep all visible text in English.