{"id":12748,"date":"2026-11-17T11:01:39","date_gmt":"2026-11-17T02:01:39","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/?p=12748"},"modified":"2026-11-17T11:10:55","modified_gmt":"2026-11-17T02:10:55","slug":"comfy-org-fibo-2-0-comfyui","status":"publish","type":"post","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/comfy-org-fibo-2-0-comfyui\/","title":{"rendered":"Fibo-2.0 Image Generation and Editing Diffusion Model: 32GB+ VRAM"},"content":{"rendered":"<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\/Comfy-Org\/Fibo-2.0\">Comfy-Org\/Fibo-2.0<\/a><\/td>\n<\/tr>\n<tr>\n<td>Family guide<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/model-briaai-fibo-2-0-dev-en\/\">Fibo-2.0-dev guide (3 articles)<\/a><\/td>\n<\/tr>\n<tr>\n<td>Published<\/td>\n<td>2026-10-21<\/td>\n<\/tr>\n<tr>\n<td>License<\/td>\n<td>cc-by-nc-4.0<\/td>\n<\/tr>\n<tr>\n<td>Formats<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/format-safetensors-en\/\">safetensors<\/a><\/td>\n<\/tr>\n<tr>\n<td>Paper<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2511.06876\">arXiv:2511.06876<\/a><\/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 when the information was collected. Dates are JST.<\/em><\/p>\n<p><!-- \/lmw:facts --><\/p>\n<h2>Overview<\/h2>\n<p>Comfy-Org has released &#8220;Comfy-Org\/Fibo-2.0&#8221;, a ComfyUI-repackaged version of the image generation and editing model &#8220;Fibo 2.0&#8221; developed by briaai. This model features a diffusion model architecture designed to perform text-to-image generation, image-to-image conversion, inpainting, and editing based on multiple reference images.<\/p>\n<p>The weight files for the original model &#8220;briaai\/Fibo-2.0-dev&#8221; and the editing-focused distillation model &#8220;briaai\/Fibo-2.0-edit-turbo&#8221; have been restructured into a single file structure and layout that is easy to handle in ComfyUI, and are provided in bf16 precision and int8 quantized versions.<\/p>\n<h2>Specifications<\/h2>\n<p>The various specifications and details of the changes that can be confirmed from the documentation are as follows:<\/p>\n<ul>\n<li><strong>Base Model<\/strong>: <code>briaai\/Fibo-2.0-dev<\/code> and <code>briaai\/Fibo-2.0-edit-turbo<\/code><\/li>\n<li><strong>Architecture and Structural Adjustments<\/strong>:<\/li>\n<li><code>bf16<\/code> version: The diffuser transformer shards are consolidated into a single file. Key names have been renamed to match ComfyUI&#8217;s structure. &#8211; The weights for attention <code>to_q<\/code>, <code>to_k<\/code>, and <code>to_v<\/code> are combined into <code>qkv<\/code>, and the <code>to_k<\/code> and <code>to_v<\/code> for cross-attention and Perceiver are combined into <code>to_kv<\/code>. The weight values themselves remain unchanged as bf16. &#8211; <code>w1<\/code> and <code>w3<\/code> are fused into <code>w13<\/code>, and the hidden width of the FFN (Feed-Forward Network) is zero-padded to 9728.<\/li>\n<li><strong>Quantization Specifications (<code>int8_convrot.safetensors<\/code>)<\/strong>:<\/li>\n<li>Weights are quantized to int8 using Hadamard rotation and per-channel scale application. &#8211; The adaLN modulation layers maintain bf16 precision.<\/li>\n<li><strong>File Placement Structure<\/strong>:<\/li>\n<li>Diffusion models (placed in <code>models\/diffusion_models\/<\/code>): &#8211; <code>fibo2_dev_bf16.safetensors<\/code> &#8211; <code>fibo2_dev_int8_convrot.safetensors<\/code> &#8211; <code>fibo2_edit_turbo_bf16.safetensors<\/code> &#8211; <code>fibo2_edit_turbo_int8_convrot.safetensors<\/code> &#8211; Text encoders (placed in <code>models\/text_encoders\/<\/code>): &#8211; <code>qwen3vl_4b_bf16.safetensors<\/code> &#8211; <code>qwen3vl_4b_int8_convrot.safetensors<\/code> &#8211; VAE (placed in <code>models\/vae\/<\/code>): &#8211; <code>flux2-vae.safetensors<\/code><\/li>\n<li><strong>License Conditions<\/strong>: The repackaged repository specifies <code>license:cc-by-nc-4.0<\/code> (Creative Commons Attribution-NonCommercial 4.0 International). It is intended for non-commercial use (note that the original model&#8217;s license is listed as <code>bria-fibo<\/code>).<\/li>\n<\/ul>\n<h2>Performance and Quality<\/h2>\n<p>While the model cards for the repackaged and original models do not provide specific comparative benchmark score tables, the published explanations and related papers (arXiv: 2511.06876) report characteristics regarding the model&#8217;s performance and quality.<\/p>\n<p>The original model, Fibo 2.0 (FIBO), is trained using long structured captions. Detailed attribute sets are uniformly assigned to each training sample, providing decoupled controllability over visual elements and advanced expressive power. In addition, it is explained that the model achieves superior prompt alignment in open-source models by introducing the &#8220;DimFusion&#8221; mechanism, which integrates intermediate tokens of lightweight LLMs without increasing token length, and &#8220;TaBR&#8221; (Text-as-a-Bottleneck Reconstruction), an evaluation protocol for measuring prompt reproducibility and controllability.<\/p>\n<p>Note that for the int8-quantized model for ComfyUI (<code>int8_convrot.safetensors<\/code>), the publisher explicitly states that the output results differ slightly from the original version in exchange for improved processing efficiency.<\/p>\n<h2>Strengths and Use Cases<\/h2>\n<p>This model, &#8220;Comfy-Org\/Fibo-2.0&#8221;, and the original Fibo 2.0 series excel at highly controllable image generation and advanced image editing tasks based on detailed attribute specifications and long structured prompts.<\/p>\n<p>As a performance characteristic of the original model, it supports not only high-quality text-to-image generation but also image-to-image conversion based on input images, inpainting for modifying specified regions, and multi-reference editing utilizing multiple input reference images. Because consistent detailed attribute annotations are applied to each training data point, generation is possible while individually controlling visual elements within the image. It is suited for creative production and professional image generation applications where precise control over intended compositions and detailed depiction is desired.<\/p>\n<p>Additionally, this repackage includes &#8220;Fibo-2.0-edit-turbo&#8221;, an editing model utilizing distillation technology, alongside the standard development model &#8220;Fibo-2.0-dev&#8221;. This makes it possible to execute interactive image editing and inpainting at high speeds with fewer inference steps.<\/p>\n<p>In operation on ComfyUI, because it is provided in a single consolidated file format (<code>safetensors<\/code>) from the original distributed structure, it contributes to simplifying node configurations and management. Furthermore, a bf16 precision version prioritizing precision and fidelity, and an int8 quantized version (int8_convrot) optimized using Hadamard rotation, are provided. Flexibility in model selection is ensured according to the application and required standards of generation quality.<\/p>\n<p><!-- lmw:hardware --><\/p>\n<h2>Hardware Requirements<\/h2>\n<p><strong>Estimated requirements (calculated by Local Model Watch)<\/strong> \u2014 8.1B parameters (taken from the base model briaai\/Fibo-2.0-edit-turbo)<\/p>\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>Your VRAM<\/th>\n<th>Quantization<\/th>\n<th>File size<\/th>\n<th>Est. memory needed<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>32GB (RTX 5090, etc.)<\/td>\n<td>BF16<\/td>\n<td>23.7GB<\/td>\n<td>28.5GB<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model&#8217;s authors. Compare with other models in our <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/vram-guide-en\/\">VRAM quick reference<\/a>. What the quantization names mean: <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/glossary-quantization-en\/\">glossary<\/a>.<\/em><\/p>\n<p><!-- \/lmw:hardware --><\/p>\n<p><!-- lmw:runnability --><\/p>\n<h2>Can You Run It Locally?<\/h2>\n<p>The publisher distributes this model as safetensors.<\/p>\n<p><strong>License \u2014 <code>cc-by-nc-4.0<\/code> (Commercial use prohibited):<\/strong> <strong>Commercial use is prohibited<\/strong> (NC = NonCommercial). Internal business use can also count as commercial, so avoid this license for work use.<\/p>\n<p><em>Compiled by this site&#8217;s code from the published formats and the license field. License summaries are not legal advice \u2014 check the publisher&#8217;s original terms before relying on them.<\/em><\/p>\n<p><!-- \/lmw:runnability --><\/p>\n<p><!-- lmw:lab --><\/p>\n<h2>Our Own Measurements<\/h2>\n<h3>Measurements We Did Not Take<\/h3>\n<p>This model&#8217;s license (<code>cc-by-nc-4.0<\/code>) does not permit commercial use. Because this site carries advertising, we did not run the model (no CPU run, answers, quantization comparison, conversion or generation).<\/p>\n<p><!-- \/lmw:lab --><\/p>\n<p><!-- lmw:files --><\/p>\n<h2>Distributed Files<\/h2>\n<p><em>Weight files published in <a href=\"https:\/\/huggingface.co\/Comfy-Org\/Fibo-2.0\/tree\/main\">Comfy-Org\/Fibo-2.0<\/a>, listed by this site from the Hugging Face API. Sizes are the actual file sizes.<\/em><\/p>\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>File<\/th>\n<th>Size<\/th>\n<th>ComfyUI folder<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>diffusion_models\/fibo2_dev_bf16.safetensors<\/code><\/td>\n<td>16.45GB<\/td>\n<td><code>models\/diffusion_models\/<\/code><\/td>\n<\/tr>\n<tr>\n<td><code>diffusion_models\/fibo2_dev_int8_convrot.safetensors<\/code><\/td>\n<td>8.41GB<\/td>\n<td><code>models\/diffusion_models\/<\/code><\/td>\n<\/tr>\n<tr>\n<td><code>diffusion_models\/fibo2_edit_turbo_bf16.safetensors<\/code><\/td>\n<td>16.45GB<\/td>\n<td><code>models\/diffusion_models\/<\/code><\/td>\n<\/tr>\n<tr>\n<td><code>diffusion_models\/fibo2_edit_turbo_int8_convrot.safetensors<\/code><\/td>\n<td>8.41GB<\/td>\n<td><code>models\/diffusion_models\/<\/code><\/td>\n<\/tr>\n<tr>\n<td><code>text_encoders\/qwen3vl_4b_bf16.safetensors<\/code><\/td>\n<td>8.88GB<\/td>\n<td><code>models\/text_encoders\/<\/code><\/td>\n<\/tr>\n<tr>\n<td><code>text_encoders\/qwen3vl_4b_int8_convrot.safetensors<\/code><\/td>\n<td>4.86GB<\/td>\n<td><code>models\/text_encoders\/<\/code><\/td>\n<\/tr>\n<tr>\n<td><code>vae\/flux2-vae.safetensors<\/code><\/td>\n<td>336MB<\/td>\n<td><code>models\/vae\/<\/code><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>The ComfyUI folder is shown only where the repository&#8217;s layout mirrors ComfyUI&#8217;s <code>models\/<\/code> directory.<\/em><\/p>\n<p><!-- \/lmw:files --><\/p>\n<h2>How to Get It<\/h2>\n<p>Model weights for this repackaged version are available in safetensors format from the <code>Comfy-Org\/Fibo-2.0<\/code> repository on Hugging Face. The models are distributed without requiring license agreements or access request procedures.<\/p>\n<p>When using this model in a ComfyUI environment, obtain the necessary safetensors files from the repository and place them in the directories corresponding to their respective roles. The specific layout structure for placement is as follows:<\/p>\n<ul>\n<li>Diffusion model placement (<code>ComfyUI\/models\/diffusion_models\/<\/code>):<\/li>\n<li><code>fibo2_dev_bf16.safetensors<\/code> &#8211; <code>fibo2_dev_int8_convrot.safetensors<\/code> &#8211; <code>fibo2_edit_turbo_bf16.safetensors<\/code> &#8211; <code>fibo2_edit_turbo_int8_convrot.safetensors<\/code><\/li>\n<li>Text encoder placement (<code>ComfyUI\/models\/text_encoders\/<\/code>):<\/li>\n<li><code>qwen3vl_4b_bf16.safetensors<\/code> &#8211; <code>qwen3vl_4b_int8_convrot.safetensors<\/code><\/li>\n<li>VAE placement (<code>ComfyUI\/models\/vae\/<\/code>):<\/li>\n<li><code>flux2-vae.safetensors<\/code><\/li>\n<\/ul>\n<p>At runtime, image generation and editing workflows can be built by connecting the text encoder (based on Qwen3-VL 4B), the diffusion model, and the FLUX.2 VAE (<code>flux2-vae<\/code>) using ComfyUI&#8217;s standard node groups. Please use combinations of the bf16 or int8 quantized files according to the required precision and operating conditions.<\/p>\n<p><!-- lmw:related --><\/p>\n<h2>Related Articles<\/h2>\n<ul>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/briaai-fibo-2-0-edit-turbo\/\">Fibo-2.0-edit-turbo Image Editing Model: 32GB+ VRAM, File List<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/bria-ai-releases-fibo-2-0-dev\/\">Fibo-2.0-dev Image Generation Model: 32GB+ VRAM, File List<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/comfy-org-releases-world-model-pan2-for-comfyui\/\">PAN2 World Model for Simulating Physical Changes: ComfyUI Paths<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/comfy-org-minimax-h3-image-released\/\">MiniMax-H3-Image 16-Megapixel Image Generation Model: 4GB+ VRAM<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:related --><\/p>\n<p><!-- lmw:next-steps --><\/p>\n<h2>What to Read Next<\/h2>\n<ul>\n<li><strong>Find models by VRAM<\/strong> (This model needs at least 32GB) \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/vram-guide-en\/\">VRAM quick reference<\/a><\/li>\n<li><strong>Explore the same model family<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/model-briaai-fibo-2-0-dev-en\/\">Fibo-2.0-dev family overview (3 articles)<\/a><\/li>\n<li><strong>Engines that run this model<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-comfyui-en\/\">ComfyUI<\/a><\/li>\n<li><strong>Formats this model is available in<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/format-safetensors-en\/\">Safetensors format guide and models<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:next-steps --><\/p>\n<h2>Sources<\/h2>\n<ul>\n<li><a href=\"https:\/\/huggingface.co\/Comfy-Org\/Fibo-2.0\">https:\/\/huggingface.co\/Comfy-Org\/Fibo-2.0<\/a><\/li>\n<li><a href=\"https:\/\/huggingface.co\/briaai\/Fibo-2.0-dev\">https:\/\/huggingface.co\/briaai\/Fibo-2.0-dev<\/a><\/li>\n<li><a href=\"https:\/\/huggingface.co\/briaai\/Fibo-2.0-edit-turbo\">https:\/\/huggingface.co\/briaai\/Fibo-2.0-edit-turbo<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2511.06876\">https:\/\/arxiv.org\/abs\/2511.06876<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Comfy-Org has released Comfy-Org\/Fibo-2.0, a ComfyUI repackage of briaai&#8217;s Fibo 2.0 image generation and editing model in bf16 and int8 formats.<\/p>\n","protected":false},"author":1,"featured_media":12747,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[203],"tags":[902,3889,1142,677,3891,3893,2089,1547,753],"class_list":["post-12748","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-image-video-and-audio","tag-bf16-en","tag-briaai-en","tag-comfy-org-en","tag-comfyui-en","tag-fibo-2-0-en","tag-fibo-2-0-dev-en","tag-int8-en","tag-verified","tag--en"],"lang":"en","translations":{"en":12748,"ja":12746},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/12748","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=12748"}],"version-history":[{"count":2,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/12748\/revisions"}],"predecessor-version":[{"id":17893,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/12748\/revisions\/17893"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media\/12747"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=12748"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/categories?post=12748"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/tags?post=12748"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}