{"id":5351,"date":"2026-03-04T21:29:51","date_gmt":"2026-03-05T02:29:51","guid":{"rendered":"https:\/\/dft.wiki\/?p=5351"},"modified":"2026-03-10T17:46:42","modified_gmt":"2026-03-10T21:46:42","slug":"generating-images-with-llm","status":"publish","type":"post","link":"https:\/\/dft.wiki\/?p=5351","title":{"rendered":"Image Manipulation with LLMs"},"content":{"rendered":"<p>Depending on the inputs a model is trained and configured to condition on, its specialities will be defined as:<\/p>\n<ul>\n<li><strong>Text-to-Image<\/strong>\n<ul>\n<li>Generate images from written prompts.<\/li>\n<li>See post [<a href=\"https:\/\/dft.wiki\/?p=5378\">Link<\/a>].<\/li>\n<\/ul>\n<\/li>\n<li><strong>Image-to-Image<\/strong>\n<ul>\n<li>Transform an existing image into a new version.<\/li>\n<li>See post [<a href=\"https:\/\/dft.wiki\/?p=5379\">Link<\/a>].<\/li>\n<\/ul>\n<\/li>\n<li><strong>Inpainting \/ Outpainting<\/strong>\n<ul>\n<li>Edit part of an image by selecting a specific area.<\/li>\n<li>See post [<a href=\"https:\/\/dft.wiki\/?p=5380\">Link<\/a>].<\/li>\n<\/ul>\n<\/li>\n<li><strong>Image Upscaling \/ Enhancing<\/strong>\n<ul>\n<li>Improve the resolution and details of low-quality images.<\/li>\n<li>See post [<a href=\"https:\/\/dft.wiki\/?p=5392\">Link<\/a>].<\/li>\n<\/ul>\n<\/li>\n<li>Other specialities\n<ul>\n<li><strong>Style Transfer<\/strong><\/li>\n<li><strong>Control-Based Generation<\/strong><\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<hr \/>\n<p><strong>TRAINING AND INFERENCE<\/strong><\/p>\n<p>Diffusion is a technique for systematically destroying data and then reconstructing it.<\/p>\n<p>It begins with a &#8220;Forward Process&#8221; where Gaussian noise is added to a clean image in successive steps until it becomes pure static (from left to right).<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-5353\" src=\"https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/diffusion.png\" alt=\"\" width=\"1242\" height=\"269\" srcset=\"https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/diffusion.png 1242w, https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/diffusion-300x65.png 300w, https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/diffusion-1024x222.png 1024w, https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/diffusion-768x166.png 768w\" sizes=\"auto, (max-width: 1242px) 100vw, 1242px\" \/><\/p>\n<p>Inference (aka &#8220;denoise&#8221;) is when a model is then tasked with the &#8220;Reverse Process&#8221; guided by a text prompt, and tries to predict exactly which &#8220;pixels&#8221; of noise were added at that specific step.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-5354\" src=\"https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/denoise.png\" alt=\"\" width=\"1242\" height=\"269\" srcset=\"https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/denoise.png 1242w, https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/denoise-300x65.png 300w, https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/denoise-1024x222.png 1024w, https:\/\/dft.wiki\/wp-content\/uploads\/sites\/15\/2026\/03\/denoise-768x166.png 768w\" sizes=\"auto, (max-width: 1242px) 100vw, 1242px\" \/><\/p>\n<p>Generated images can be quite impressive. Check out my <strong>AI Gallery<\/strong> at [<a href=\"https:\/\/ai-gallery.dft.wiki\/\">Link<\/a>].<\/p>\n<hr \/>\n<p><strong>READ MORE<\/strong><\/p>\n<ul>\n<li>Acronyms, Jargon, and Architecture of LLM and Generative AI [<a href=\"https:\/\/dft.wiki\/?p=5347\">Link<\/a>].<\/li>\n<li>Interacting Directly with Ollama\u2019s API [<a href=\"https:\/\/dft.wiki\/?p=5308\">Link<\/a>].<\/li>\n<li>Self-hosted AI Models for Coding and More [<a href=\"https:\/\/dft.wiki\/?p=5270\">Link<\/a>].<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Depending on the inputs a model is trained and configured to condition on, its specialities [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[12],"tags":[],"class_list":["post-5351","post","type-post","status-publish","format-standard","hentry","category-ai"],"_links":{"self":[{"href":"https:\/\/dft.wiki\/index.php?rest_route=\/wp\/v2\/posts\/5351","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dft.wiki\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dft.wiki\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dft.wiki\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/dft.wiki\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5351"}],"version-history":[{"count":18,"href":"https:\/\/dft.wiki\/index.php?rest_route=\/wp\/v2\/posts\/5351\/revisions"}],"predecessor-version":[{"id":5393,"href":"https:\/\/dft.wiki\/index.php?rest_route=\/wp\/v2\/posts\/5351\/revisions\/5393"}],"wp:attachment":[{"href":"https:\/\/dft.wiki\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5351"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dft.wiki\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5351"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dft.wiki\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5351"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}