圖像生成🐭

生成對抗網絡(GANs) 或 擴散模型(Diffusion model) 是常見的圖像生成技術，透過訓練模型來學習大量圖像數據的分佈，從而生成新的圖像。當輸入一個prompt時，模型會將這個prompt轉換為向量表示，並使用這些向量來引導圖像生成的過程，使生成的圖像與prompt的內容和風格相符。

在電腦視覺中，圖像可以轉換成矩陣表示，這些矩陣可以用於進行similarity search，幫助找到與輸入prompt或其他圖像相似的內容，以進行進一步的修改或應用。

圖像生成GenAI🐮

DALL‧E

DALL·E是由OpenAI開發的圖像生成工具，使用者可以透過ChatGPT的Plugin或串接API來生成圖像。

Prompt：

A futuristic basketball arena with a sleek, high-tech design, glowing neon lights, and holographic displays. The starting five players of a basketball team are making an epic entrance onto the court. Each player is wearing advanced, sci-fi-inspired uniforms with glowing accents and advanced gear. The court itself is made of a translucent, illuminated material, with dynamic graphics moving across the floor. The players are in dynamic, powerful poses, exuding confidence and energy as they prepare for the game, with the crowd's excitement palpable in the background.

圖像生成結果：

Midjourney

Midjourney是一款在Discord平台上運行的圖像生成工具，儘管目前沒有免費版，透過在Discord上輸入prompt，Midjourney可以快速生成富有藝術感的圖像。

Midjourney圖像生成作品：

Midjourney Discord畫面：

Leonardo.AI

Leonardo.AI是一款圖像生成工具，每天提供一定額度的tokens供使用者進行圖像生成。不同類型的生成內容消耗的tokens數量不同，使用者可以根據需求選擇合適的生成方式。

Leonardo.AI主頁：

Prompt：

In the final seconds of an intense basketball game, the scene captures the star player in a breathtaking moment of athleticism and determination. The player, wearing a jersey with bold team colors, is airborne, fully extended as they release the ball just as the game clock hits zero. The scoreboard above shows a tie game, with the buzzer about to sound, adding to the tension of the moment. Surrounding the player are three fierce defenders, each in mid-jump, arms outstretched, desperately trying to block the shot. Their faces show a mix of focus and desperation, knowing that this shot could decide the game. The star player's expression is one of pure concentration, eyes locked on the hoop, unfazed by the defensive pressure.

The basketball court beneath them is polished to a shine, reflecting the bright overhead lights. The crowd in the background is a blur of motion, with some fans on their feet, hands in the air, while others hold their breath in anticipation. The arena is electric with energy, the tension palpable, as everyone watches the ball arcing towards the basket. The moment is frozen in time, capturing the drama, skill, and high stakes of this buzzer-beater shot that could win the game.

圖像生成結果：

總結

這些工具各有其特點：DALL·E擅長處理複雜場景的生成、Midjourney適合藝術性強的圖像創作，而Leonardo.AI則提供了靈活的token系統，讓使用者能夠依需求生成圖像。值得注意的是，使用英文來撰寫prompt通常能生成更符合預期的圖片，因為這些圖像生成模型大多是在大量的英文數據上訓練的，因此對英文的理解更為深刻和準確。

題外話😂

Meta宣布要在2025年下架FB和IG中的非官方濾鏡，我個人覺得這又是一波斂財操作。順帶一提，如果你想保留之前用濾鏡拍的reels或限時動態，記得去把影片下載起來。

參考資料💯

• Midjourney
• Leonardo.A