
AI & computing
Diffusion model: forward noising and reverse denoising
- Layout:
- one row of images with opposite arrows above and below
- Aspect ratio:
- 16:9
Prompt
Create a schematic of 【a diffusion model for image generation】: one horizontal row of five small images. The leftmost is a clear picture of 【a flower】; moving right, each image has more speckled noise, and the rightmost is pure gray static.
1. Above the row, one long arrow points from left to right: the forward process adds noise step by step.
2. Below the row, one long arrow points from right to left: the reverse process removes noise step by step.
3. The lower arrow passes through one small box for the denoising network, placed under the middle of the row.
Only the leftmost and rightmost images get labels.
Write each label only once in the whole figure.
Labels, each used once, no other text: 【Clean image, Pure noise, Forward process: add noise, Reverse process: denoise, Denoising network】.
Style: journal-style schematic diagram. Clean flat vector graphics on a white background. Modules are rounded rectangles or simple icons with their short names inside; light pastel panels group related layers. Only 3-4 muted colors, each with one fixed meaning. Thin, uniform arrows show the flow of data, energy, material or influence; dashed arrows for indirect or feedback links. Sans-serif English labels only, each label appears exactly once, spelled exactly as given. No 3D rendering, no shadows, no glow, no cyberpunk or tech-background decoration, no photorealism.
Opens Yila with this prompt and the example as a reference image.
The examples were generated with Nano Banana. The prompt also works with GPT Image and other image models; results vary by model.






