{"id":5379,"date":"2026-03-10T07:10:05","date_gmt":"2026-03-10T11:10:05","guid":{"rendered":"https:\/\/dft.wiki\/?p=5379"},"modified":"2026-03-10T20:51:20","modified_gmt":"2026-03-11T00:51:20","slug":"ai-image-transformation-image-to-image","status":"publish","type":"post","link":"https:\/\/dft.wiki\/?p=5379","title":{"rendered":"AI Image Transformation – Image-to-Image"},"content":{"rendered":"

IMAGE-TO-IMAGE<\/strong><\/p>\n

While Text-to-Image<\/strong> starts the inference process from absolute random noise, Image-to-Image<\/strong> partially diffuses an existing image and then denoises it to introduce new features based on a prompt.<\/p>\n

Since the pipeline receives an image as its initial input, the file must be loaded into memory before processing.<\/p>\n

from PIL import Image\r\nfrom diffusers import StableDiffusionImg2ImgPipeline\r\n\r\npipe = StableDiffusionImg2ImgPipeline.from_pretrained(\r\n    \"stable-diffusion-v1-5\/stable-diffusion-v1-5\"<\/strong>,\r\n)\r\n\r\nresult = pipe(\r\n    prompt=\"Smiling face.<\/strong>\",\r\n    image=Image.open(\"flower.png\").convert(\"RGB\")<\/strong>,\r\n    strength=0.5<\/strong>,\r\n    guidance_scale=7.5<\/strong>,\r\n).images[0]\r\n\r\nresult.save(\"output.png<\/strong>\")<\/pre>\n
Note the new parameter: strength<\/strong>. It controls the amount of noise added to your input image before the denoising process begins. High strength values allow for more creative freedom, while low values stay closer to the original.<\/p>\n
Example:<\/strong><\/p>\n