Diffusion Model
A diffusion model is a type of generative AI that turns random noise into an image by removing that noise step by step, guided by a prompt.
June 16, 2026
A diffusion model is a generative AI system that creates images by starting from pure random noise and progressively removing it, step by step, until a coherent picture emerges.
How it works
A diffusion model is trained in two directions. During training, it takes real images and adds random noise to them in many small increments until each image becomes pure static — this is the "forward" process. The model learns to predict and reverse each of those noising steps.
To generate a new image, it runs that learned process backward. It begins with a field of random noise and, across dozens of small steps, predicts what noise to remove at each stage. A text prompt steers every step, nudging the emerging picture toward the words you typed. After the final step, the noise has been transformed into a clean, original image. Most modern systems run this process in a compressed "latent space" for speed, then decode the result into full-resolution pixels.
Why it matters
Diffusion replaced earlier approaches like GANs as the dominant method for AI image generation because it is more stable to train and produces a wider range of high-quality, detailed results. It follows prompts faithfully, handles complex scenes well, and gives creators fine control through settings such as the seed, CFG scale, and number of denoising steps. The same framework now powers AI video, where diffusion runs across frames to keep motion smooth and consistent.
In eaxy
eaxy uses diffusion-based generation under the hood so you never touch the math — you describe what you want, pick a style pack, and the model handles the denoising. From there you can animate stills into motion with Kling 3 and export up to 4K, all from plain-language prompts.
Related terms
Frequently asked questions
What is a diffusion model in simple terms?+
It is an AI that learns to reverse a noising process. It starts with random static and, over many small steps, cleans it up into a clear image that matches your prompt.
Are diffusion models the same as GANs?+
No. Both generate images, but a GAN uses two competing networks in one pass, while a diffusion model refines an image gradually across many denoising steps, which tends to give more stable, varied results.
Why are diffusion models so popular for image generation?+
They produce high-quality, diverse images and follow text prompts closely. Most modern text-to-image tools, including Stable Diffusion, are built on diffusion.
Do diffusion models also make video?+
Yes. Video diffusion extends the same idea across frames so motion stays consistent, which is how modern AI video models generate clips.
Make it with eaxy
Describe anything and generate stunning images in seconds — then bring them to motion with Kling 3.