{"id":24883,"date":"2026-08-21T18:43:35","date_gmt":"2026-08-21T18:43:35","guid":{"rendered":"https:\/\/jurn.link\/dazposer\/?p=24883"},"modified":"2026-08-21T18:44:43","modified_gmt":"2026-08-21T18:44:43","slug":"minimax-h3-progress","status":"publish","type":"post","link":"https:\/\/jurn.link\/dazposer\/index.php\/2026\/08\/21\/minimax-h3-progress\/","title":{"rendered":"Minimax H3 progress"},"content":{"rendered":"<p>I&#8217;m getting much better results with Minimax, and little quicker, since I moved to the latest 1.1 version of the most popular 4-step turbo LoRA.<\/p>\n<p>I&#8217;m now using 32Gb of files in total on a 12Gb VRAM card, an incredible feat made possible by the behind-the-scenes magic of Minimax, ComfyUI and the .W4A8 format (successor to GGUFs)&#8230;<\/p>\n<p><strong>Model:<\/strong> <em>minimax_h3_ref2va_hybrid_b20-49_pruned_w4a8_mixed.safetensors<\/em> (11.6Gb) <\/p>\n<p><strong>Text encoder:<\/strong> <em>qwen3vl_32b_minimax_h3-w4a8_convrot.safetensors<\/em> (14.6Gb)<\/p>\n<p><strong>Video VAE:<\/strong> <em>minimax_h3_video_vae_int8_convrot.safetensors<\/em> (2.9Gb) (still the good old fast one)<\/p>\n<p><strong>Audio VAE:<\/strong> still the official one (577Mb)<\/p>\n<p><strong>Turbo LoRA:<\/strong> <em>minimax_h3_fl2v_turbo_4step_v1.1_768p_comfyui_bf16.safetensors<\/em> (1.8Gb)<\/p>\n<p>Generating in 4 steps, <em>er_sde<\/em> with <em>sgm_uniform<\/em> for visual quality.<\/p>\n<p>The audio is bad, due to the turbo 4-steps and the <em>er-sde<\/em> sampler, but audio would be replaced manually.  Tests initially suggest the turbo LoRA should be left at 1.0, regardless of video size. <\/p>\n<p><strong>Tests, 10 seconds:<\/strong><\/p>\n<p>0.3 megapixels.<\/p>\n<div style=\"width: 640px;\" class=\"wp-video\"><video class=\"wp-video-shortcode\" id=\"video-24883-1\" width=\"640\" height=\"362\" preload=\"metadata\" controls=\"controls\"><source type=\"video\/mp4\" src=\"https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/08\/Minimax_00011_.mp4?_=1\" \/><a href=\"https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/08\/Minimax_00011_.mp4\">https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/08\/Minimax_00011_.mp4<\/a><\/video><\/div>\n<p>&nbsp;<\/p>\n<p>0.5 megapixels.<\/p>\n<div style=\"width: 640px;\" class=\"wp-video\"><video class=\"wp-video-shortcode\" id=\"video-24883-2\" width=\"640\" height=\"363\" preload=\"metadata\" controls=\"controls\"><source type=\"video\/mp4\" src=\"https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/08\/Minimax_00006_.mp4?_=2\" \/><a href=\"https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/08\/Minimax_00006_.mp4\">https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/08\/Minimax_00006_.mp4<\/a><\/video><\/div>\n<p>&nbsp;<\/p>\n<p>1.0 megapixels.<\/p>\n<div style=\"width: 640px;\" class=\"wp-video\"><video class=\"wp-video-shortcode\" id=\"video-24883-3\" width=\"640\" height=\"357\" preload=\"metadata\" controls=\"controls\"><source type=\"video\/mp4\" src=\"https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/08\/Minimax_00007_.mp4?_=3\" \/><a href=\"https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/08\/Minimax_00007_.mp4\">https:\/\/jurn.link\/dazposer\/wp-content\/uploads\/2026\/08\/Minimax_00007_.mp4<\/a><\/video><\/div>\n<p>&nbsp;<\/p>\n<p>As you can see, much of the graininess on the complex fast-moving net and dragonfly wings goes away at the native 1.0. Not completely, but apparently such problems are vanquished as one nears  2.0.<\/p>\n<p>I&#8217;m still testing and building the text-to-video workflow, and have not yet started using reference images which would fix the owl&#8217;s visual character etc. Also, the next version of ComfyUI Portable will add the ability to include keyframes anywhere, which will make generation less of a slot-machine and more of an assembly tool.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>I&#8217;m getting much better results with Minimax, and little quicker, since I moved to the latest 1.1 version of the most popular 4-step turbo LoRA. I&#8217;m now using 32Gb of files in total on a 12Gb VRAM card, an incredible feat made possible by the behind-the-scenes magic of Minimax, ComfyUI and the .W4A8 format (successor [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7,12],"tags":[],"class_list":["post-24883","post","type-post","status-publish","format-standard","hentry","category-the-animation-industry","category-tutorials"],"_links":{"self":[{"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/posts\/24883","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=24883"}],"version-history":[{"count":2,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/posts\/24883\/revisions"}],"predecessor-version":[{"id":24888,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/posts\/24883\/revisions\/24888"}],"wp:attachment":[{"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/media?parent=24883"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/categories?post=24883"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jurn.link\/dazposer\/index.php\/wp-json\/wp\/v2\/tags?post=24883"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}