{"id":2546,"date":"2026-09-23T06:12:09","date_gmt":"2026-09-22T21:12:09","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/2026\/09\/23\/supra2-img-lightweight-104m-text-to-image-model\/"},"modified":"2026-09-24T17:10:52","modified_gmt":"2026-09-24T08:10:52","slug":"supra2-img-lightweight-104m-text-to-image-model","status":"publish","type":"post","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/09\/23\/supra2-img-lightweight-104m-text-to-image-model\/","title":{"rendered":"Supra2-IMG: A Lightweight 104M Text-to-Image Model"},"content":{"rendered":"<p><em>Sample outputs are available on the <a href=\"https:\/\/huggingface.co\/SupraLabs\/Supra2-IMG\">model card<\/a>.<\/em><\/p>\n<p><!-- lmw:facts --><\/p>\n<h2>At a Glance<\/h2>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Item<\/th>\n<th>Value<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Repository<\/td>\n<td><a href=\"https:\/\/huggingface.co\/SupraLabs\/Supra2-IMG\">SupraLabs\/Supra2-IMG<\/a><\/td>\n<\/tr>\n<tr>\n<td>Published<\/td>\n<td>2026-09-21<\/td>\n<\/tr>\n<tr>\n<td>License<\/td>\n<td>apache-2.0<\/td>\n<\/tr>\n<tr>\n<td>Source type<\/td>\n<td>Primary source (the publisher itself)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Values determined by this site&#8217;s code at collection time. Dates are JST.<\/em><\/p>\n<p><!-- \/lmw:facts --><\/p>\n<p><em>Sample outputs are available on the <a href=\"https:\/\/huggingface.co\/SupraLabs\/Supra2-IMG\">model card<\/a>.<\/em><\/p>\n<p><em>Sample outputs are available on the <a href=\"https:\/\/huggingface.co\/SupraLabs\/Supra2-IMG\">model card<\/a>.<\/em><\/p>\n<p><em>Sample outputs are available on the <a href=\"https:\/\/huggingface.co\/SupraLabs\/Supra2-IMG\">model card<\/a>.<\/em><\/p>\n<h2>Overview<\/h2>\n<p>SupraLabs has released &#8220;Supra2-IMG&#8221;, an extremely lightweight text-to-image generation model trained from scratch using high-quality synthetic datasets. This model is designed to provide state-of-the-art (SOTA) image quality for its scale, despite having a very small size of approximately 100M (100 million) parameters.<\/p>\n<h2>Specifications<\/h2>\n<p>The main specifications and architecture of the model are as follows:<\/p>\n<ul>\n<li>Architecture: Diffusion Transformer (DiT) \/ SupraDiT<\/li>\n<li>Parameter count: 104.1M<\/li>\n<li>Input: Text<\/li>\n<li>Output resolution: 256\u00b2<\/li>\n<li>Latent size: 32\u00b2<\/li>\n<li>Patch size: 2<\/li>\n<li>Text encoder: <a href=\"https:\/\/huggingface.co\/google\/flan-t5-base\">Flan-T5-Base<\/a> (frozen)<\/li>\n<li>VAE: <a href=\"https:\/\/huggingface.co\/stabilityai\/sd-vae-ft-mse\">SD-VAE-FT-MSE<\/a><\/li>\n<li>Recommended sampling settings:\n<ul>\n<li>CFG scale: 3.0 &#8211; Steps: 50 &#8211; Seed: 0<\/li>\n<\/ul>\n<\/li>\n<li>Model configuration details:\n<ul>\n<li><code>D_MODEL<\/code>: 576 &#8211; <code>DEPTH<\/code>: 14 &#8211; <code>N_HEADS<\/code>: 9 &#8211; <code>HEAD_DIM<\/code>: 64 &#8211; <code>MLP_RATIO<\/code>: 4.0 &#8211; <code>D_CTX<\/code>: 768 &#8211; <code>VAE_SCALE<\/code>: 0.18215<\/li>\n<\/ul>\n<\/li>\n<li>License: apache-2.0<\/li>\n<\/ul>\n<h2>Performance and Quality<\/h2>\n<p>Supra2-IMG was trained for 10 epochs using the &#8220;LucasFang\/FLUX-Reason-6M&#8221; dataset, which contains 5.6M (5.6 million) images. In the data preparation process, specific priorities are established for prompt selection in order to maximize the quality of the data used for training. Specifically, prompts are selected in the order of <code>caption_composition<\/code>, <code>caption_entity<\/code>, <code>caption_text<\/code>, <code>caption_style<\/code>, and <code>caption_imaginative<\/code>, functioning as a fallback when the preceding prompt is unavailable, thereby strictly managing pre-training to use only the highest quality data.<\/p>\n<p>Regarding the training environment, it is reported that training was completed in a total of 9 hours, including data preparation time, using a single Nvidia H100 SXM 80GB (Runpod Pod). Additionally, 2.5TB of disk space was used for the training.<\/p>\n<p>According to the model card, it achieves SOTA (State-of-the-art) quality at this extremely small parameter count of 104.1M. While specific numerical benchmark comparisons are not provided, it is emphasized that the combination of high-quality synthetic data and an efficient DiT architecture yields high generation capability relative to the model size.<\/p>\n<h2>Strengths and Use Cases<\/h2>\n<p>Supra2-IMG is a text-to-image (T2I) model whose greatest feature is achieving SOTA-class quality relative to its model size with an extremely small parameter count of just 104.1M. This is largely attributed to the rigorous preparation standards applied to the high-quality synthetic dataset containing 5.6M images, &#8220;LucasFang\/FLUX-Reason-6M&#8221;, during the training process.<\/p>\n<p>Specifically, a fallback method is adopted in the data preparation stage where the quality of prompts is evaluated in the order of <code>caption_composition<\/code>, <code>caption_entity<\/code>, <code>caption_text<\/code>, <code>caption_style<\/code>, and <code>caption_imaginative<\/code>, and the next prompt is used only if the quality of the previous stage is not met. Through this process, the model is trained exclusively on the highest quality data, which is expected to result in high-precision image generation even with limited computational resources and memory capacity.<\/p>\n<p>Generation is currently intended for English (en), and plans have been indicated to evolve into a next-generation model such as &#8220;Supra2.5-IMG&#8221; in the future. It will serve as a very strong option for engineers who value local execution and seek high-quality generation while remaining lightweight.<\/p>\n<h2>How to Get It<\/h2>\n<p>The model can be used via the inference script provided in the Hugging Face repository. The distribution format combines the script and the model weight file (<code>model_final_ema.pt<\/code>).<\/p>\n<p>First, run the following commands to create a working directory and obtain the inference script <code>inference.py<\/code>.<\/p>\n<pre><code class=\"language-bash\"># Create project directory\nmkdir Supra2-IMG\ncd Supra2-IMG\n\n# Download inference script\nwget https:\/\/huggingface.co\/SupraLabs\/Supra2-IMG\/resolve\/main\/inference.py\n<\/code><\/pre>\n<p>Next, you can start image generation using the <code>python inference.py<\/code> command. During execution, in addition to the <code>--prompt<\/code> argument describing what you want to generate, you can specify options such as <code>--seed<\/code> to ensure reproducibility, <code>--cfg<\/code> (recommended value: 3.0) to control generation quality, and <code>--steps<\/code> (recommended value: 50) to determine sampling precision. Additionally, <code>--n<\/code> to indicate the number of images to generate and <code>--out<\/code> to specify the output filename are also available.<\/p>\n<pre><code class=\"language-bash\">python inference.py --prompt &quot;a sea jellyfish floating in the pitch-black ocean depths&quot;  --seed 0  --cfg 3.0  --steps 50  --n 1  --out jellyfish.png\n<\/code><\/pre>\n<p>The execution logs display details such as the device being used, the loaded checkpoint, the model parameter count, the loading status of the text encoder (Flan-T5-Base), and the sampling process (Euler flow).<\/p>\n<p>Note that downloading this model does not require a license agreement (gated access) on Hugging Face.<\/p>\n<h2>Sources<\/h2>\n<ul>\n<li><a href=\"https:\/\/huggingface.co\/SupraLabs\/Supra2-IMG\">https:\/\/huggingface.co\/SupraLabs\/Supra2-IMG<\/a><\/li>\n<\/ul>\n<p><!-- lmw:updates --><\/p>\n<h2>Update History<\/h2>\n<ul>\n<li>2026-09-23: Rebuilt the article (details are in the Japanese edition).<\/li>\n<li>2026-09-23: Rebuilt the article (details are in the Japanese edition).<\/li>\n<\/ul>\n<p><!-- \/lmw:updates --><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Discover Supra2-IMG, a lightweight 104M parameter text-to-image model trained from scratch on high-quality synthetic data.<\/p>\n","protected":false},"author":1,"featured_media":2545,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[203],"tags":[1833,2039,2041,2043,1547,753],"class_list":["post-2546","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-image-video-and-audio","tag-dit-en","tag-supra2-img-en","tag-supralabs-en","tag-text-to-image-en","tag-verified","tag--en"],"lang":"en","translations":{"en":2546,"ja":2544},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/2546","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/comments?post=2546"}],"version-history":[{"count":3,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/2546\/revisions"}],"predecessor-version":[{"id":3114,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/2546\/revisions\/3114"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media\/2545"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=2546"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/categories?post=2546"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/tags?post=2546"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}