{"id":24646,"date":"2026-05-22T19:12:13","date_gmt":"2026-05-22T19:12:13","guid":{"rendered":"https:\/\/jurn.link\/dazposer\/?p=24646"},"modified":"2026-05-22T20:35:56","modified_gmt":"2026-05-22T20:35:56","slug":"how-to-install-stable-audio-3-0-small-sfx-safetensors-and-run-it-in-comfyui","status":"publish","type":"post","link":"https:\/\/jurn.link\/dazposer\/index.php\/2026\/05\/22\/how-to-install-stable-audio-3-0-small-sfx-safetensors-and-run-it-in-comfyui\/","title":{"rendered":"How to install Stable Audio 3.0 Small SFX safetensors, and run it in ComfyUI"},"content":{"rendered":"<p>How to install Stable Audio 3.0 Small SFX safetensors, and run it in ComfyUI.<\/p>\n<p>Why the Small SFX version of the new <a href=\"https:\/\/jurn.link\/dazposer\/index.php\/2026\/05\/20\/release-stable-audio-3-with-comfyui-checkpoints-and-encoder\/\">Stable Audio 3.0<\/a>? Because it&#8217;s amazingly quick, commercial-use, and also because the larger 3.0 main or base version requires Flash Attention 2. As the official readme states&#8230;<\/p>\n<blockquote><p>&#8220;Stable Audio 3 Medium requires Flash Attention 2.&#8221;<\/p><\/blockquote>\n<p>Good luck with that, then. The rest of us will use the Small models. What follows should theoretically also work for the Small Music model, but I&#8217;m mainly interested in the audio &#8216;foley&#8217; sound-effects generator model.<\/p>\n<hr>\n<p>Here&#8217;s the install guide for ComfyUI:<\/p>\n<p><strong>1.<\/strong> Upgrade your ComfyUI portable to the latest 0.22 version or higher (required, not optional) which had zero-day support for SA3 a few days ago. Then also run PIP to install the new <em>requirements.txt<\/em> as well.<\/p>\n<p><strong>2.<\/strong> From <a href=\"https:\/\/huggingface.co\/Comfy-Org\/stable-audio-3\">the official ComfyUI HuggingFace<\/a>, download the <strong>stable_audio_3_small_sfx.safetensors<\/strong> (2.3Gb) and put it your local ComfyUI&#8217;s <em>..\\models\\checkpoints<\/em> folder and also the <strong>t5gemma_b_b_ul2.safetensors<\/strong> (1.2Gb) and put that in <em>..\\models\\text_encoders<\/em> No config file is needed.<\/p>\n<p><strong>3.<\/strong> <a href=\"https:\/\/huggingface.co\/stabilityai\/stable-audio-3-small-sfx\">The official readme for SA3 Small SFX<\/a> says it needs:<\/p>\n<blockquote><p>Steps = 8.<br \/>\nCFG = 1.0.<br \/>\nSampler = Pingpong<\/p><\/blockquote>\n<p>Hmm&#8230; what, Pingpong?? Never heard of it. Turns it&#8217;s a custom node and sampler by Blepping, all in one. It&#8217;s <a href=\"https:\/\/gist.githubusercontent.com\/blepping\/b372ef6c5412080af136aad942d9d76c\/raw\/199c387b6b6aaa1dd90cc9f138fe78bda813d935\/pingpongsampler_node.py\">here as <em>pingpongsampler_node.py<\/em><\/a>. Drop this file in the root of your ComfyUI Custom Nodes folder, and re-start Comfy. It has no <em>requirements.txt<\/em>.<\/p>\n<p>Once ComfyUI is loaded with PingPong, the new sampler won&#8217;t show up on the list of samplers in your regular nodes. Instead you just double-click on a workflow and type <em>pingpong<\/em>, and then load it via its own node.<\/p>\n<p><strong>4.<\/strong> Now assemble the following ComfyUI workflow thus. This works for me and gives reasonable results with blistering speed. I say &#8220;reasonable&#8221; because I still think that <a href=\"https:\/\/archive.org\/details\/stable-audio-open-1.0-webui-portable-win\">Stable Audio 1.0<\/a> gives a better quality of output and also seems to handle the instruction to &#8216;mix&#8217; sounds better, but then&#8230; 1.0 also takes about 50 times as long to generate an audio clip. If you have a super-ninja graphics card, that may not matter much. But for the GPU-poor it may matter.<\/p>\n<p><a href=\"https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/05\/stable_audio_3-sfx-with-pingpong-workflow-demo.jpg\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/05\/stable_audio_3-sfx-with-pingpong-workflow-demo-1024x524.jpg\" alt=\"\" width=\"640\" height=\"328\" class=\"aligncenter size-large wp-image-24647\" srcset=\"https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/05\/stable_audio_3-sfx-with-pingpong-workflow-demo-1024x524.jpg 1024w, https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/05\/stable_audio_3-sfx-with-pingpong-workflow-demo-300x154.jpg 300w, https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/05\/stable_audio_3-sfx-with-pingpong-workflow-demo-768x393.jpg 768w, https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/05\/stable_audio_3-sfx-with-pingpong-workflow-demo-1536x786.jpg 1536w, https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/05\/stable_audio_3-sfx-with-pingpong-workflow-demo.jpg 1864w\" sizes=\"auto, (max-width: 640px) 100vw, 640px\" \/><\/a><\/p>\n<p>You may want to also hook an audio output to <em>Denoised output<\/em> and compare the two.<\/p>\n<p>Possibly there are going to be better ways to do it. Possibly I&#8217;m doing it wrong. But for now, in the first day or so after release, this works for me.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>How to install Stable Audio 3.0 Small SFX safetensors, and run it in ComfyUI. Why the Small SFX version of the new Stable Audio 3.0? Because it&#8217;s amazingly quick, commercial-use, and also because the larger 3.0 main or base version requires Flash Attention 2. As the official readme states&#8230; &#8220;Stable Audio 3 Medium requires Flash [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[13,12],"tags":[],"class_list":["post-24646","post","type-post","status-publish","format-standard","hentry","category-companion-software","category-tutorials"],"_links":{"self":[{"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/posts\/24646","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/comments?post=24646"}],"version-history":[{"count":11,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/posts\/24646\/revisions"}],"predecessor-version":[{"id":24658,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/posts\/24646\/revisions\/24658"}],"wp:attachment":[{"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/media?parent=24646"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/categories?post=24646"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/tags?post=24646"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}