Documento original
GPT Image 2 · zh
This is an adorable horizontal, hand-drawn scrapbook-style Chinese science infographic. The background is warm beige paper with rounded borders, macaron-colored doodles, washi tape stickers, and little stars, creating a classroom notebook atmosphere. At the top center is a striking handwritten Chinese title: "What is Tokenization?", with the latter half in coral pink and the rest in dark brown. In the upper left corner is a cute Shiba Inu teacher character, wearing a light yellow shirt and blue tie, facing forward, holding a pointer upwards, with a speech bubble next to him saying: "Today we'll learn about tokenization!" In the upper right corner is a cute blue square Token mascot, labeled "TOKEN," smiling with rosy cheeks, tiny limbs, and holding a magnifying glass; next to it is a speech bubble saying: "I am a Token! The more rationally it's segmented, the smarter the model!" Near the right edge, a small hanging note card is added, saying "Knowledge Points Collection" and accompanied by a heart icon. Below the title, a rounded dotted text box contains the definition: "Tokenization is the process of dividing human-readable text into tokens." The word "Token" is highlighted in blue. Below that, a line of explanatory text is added: "Let's take 'Programmer Left' and 'Artificial Intelligence' as examples to see different tokenization methods:" "Programmer Left" is highlighted in light blue, and "Artificial Intelligence" in pink. The lower half contains four rounded, pastel-colored panels arranged from left to right, each with a numbered badge and title. Panel 1 is mint green, titled "1-Word-Level Tokenization," with the subtitle "Segmenting Text Based on Complete Semantics," and contains three token bubbles: "Programmer," "Left," and "Artificial Intelligence." At the bottom of this panel, a Shiba Inu teacher gives a thumbs-up, with a speech bubble that reads: "Semantic complete, easy to understand!" Panel 2 is light yellow, titled "2-Character Level Token Segmentation" with the subtitle "Character-by-Character Text Segmentation." It contains nine circular token bubbles arranged in a row: "程" (program), "序" (sequence), "员" (person), "L" (L), "e" (e), "f" (f), "t" (t), "人" (person), "工" (work), "智" (intelligence), and "能" (ability)—presenting individual characters in the example in a compact grid. At the bottom, a blue token mascot holds a pencil, with a speech bubble that reads: "Fineest granularity, but the sequences will be very long~." Panel 3 is light pink, titled "3-Sub-Word Segmentation" with the subtitle "Segmentation by Root and Prefix/Suffix." It contains five token bubbles: "程序" (program), "员" (person), "Left," "人工" (artificial), and "智能" (intelligence). At the bottom, a Shiba Inu teacher is shown thinking, with a paw near its mouth and a question mark above its head. The speech bubble reads: "Compromise solution, often works very well!" Panel 4 is light blue, titled "4-Byte Level Word Segmentation," with the subtitle "Converting Words to Byte Encoding, Combining and Dividing Based on the Conversion." It displays 14 small circular code bubbles: "E7," "A8," "8B," "...", "66," "74," "E4," "BA," "BA," "E8," "83," "BD," "..." arranged in multiple rows like encoded bytes. At the bottom, a smiling blue Token mascot is displayed again, with a speech bubble that reads, "Suitable for processing any text!" Next to it is a tiny light bulb doodle. At the very bottom, a rounded banner notes, "Different word segmentation methods have their advantages; in actual applications, the appropriate solution will be chosen based on the task and data!" All four panels use soft macaron green, yellow, pink, and blue, paired with dark brown handwritten font, delicate shadows, rounded shapes, and cute expressions. The overall composition is balanced, simple, and user-friendly, like a beautifully designed social media science poster.Como usar
- Copie o prompt original completo, incluindo os parâmetros.
- Adapte o tema, o texto e a composição aos seus requisitos.
- Use uma ferramenta externa compatível; confira o acesso e o resultado.
Antes de começar
- Verifique se é necessária uma imagem de referência. Compatibilidade não comprova qual modelo gerou o exemplo.
- Confira nomes, números e textos; os resultados variam entre versões do modelo.
