<p>Boogu-Image Edit is the image editing build in the Boogu-Image 0.1 family, a 10B-parameter model that takes a text prompt plus a source image and returns an edited result. It handles object insertion, replacement, and removal, attribute changes, and text editing while keeping the rest of the scene intact.</p><p><strong>Originally released by the </strong><a target="_blank" rel="ugc" href="https://huggingface.co/Boogu"><strong>Boogu team</strong></a><strong> on Hugging Face.</strong> All credit for the model goes to the Boogu project and its contributors. Civitai is hosting a mirror so creators can run it on-site without a local setup - please head to the <a target="_blank" rel="ugc" href="https://github.com/boogu-project/Boogu-Image">original repo</a> and <a target="_blank" rel="ugc" href="https://boogu.org">project site</a> for weights, updates, and to support the project directly.</p><h3 id="built-by-the-boogu-project"><strong>Built by the Boogu project</strong></h3><ul><li><p><a target="_blank" rel="ugc" href="https://boogu.org">Boogu project</a> - model training and release</p></li><li><p><a target="_blank" rel="ugc" href="https://huggingface.co/Boogu">Hugging Face org</a> and <a target="_blank" rel="ugc" href="https://modelscope.cn/organization/Boogu">ModelScope org</a> - weights and variant downloads</p></li><li><p>Acknowledged upstream influences: Qwen-Image, Z-Image, OmniGen2, FLUX, and DeepSeek</p></li></ul><h3 id="what-it-does"><strong>What it does</strong></h3><p>Edit is a text-plus-image-to-image model for targeted changes: add or remove objects, swap attributes (color, material, style), and edit rendered text, all driven by a natural-language instruction. It shares the family's bilingual Chinese and English text handling and runs at 25 to 50 steps with CFG in the 2.0 to 5.0 range.</p><h3 id="versions-on-civitai"><strong>Versions on Civitai</strong></h3><p>We mirror the v0.1 release in three diffusion builds: <code>bf16</code> full precision (~20GB), <code>fp8_scaled</code> (~10GB), and <code>nvfp4</code> (~5.7GB). The Qwen3-VL 8B text encoder and the Flux VAE are bundled so you can assemble a complete editing workflow.</p><h3 id="related-boogu-models"><strong>Related Boogu models</strong></h3><p><a target="_blank" rel="ugc" href="/models/2714299">Boogu-Image Base</a> is the text-to-image foundation model. <a target="_blank" rel="ugc" href="/models/2714686">Boogu-Image Turbo</a> is the fast 4-step text-to-image variant.</p><h3 id="known-limitations"><strong>Known limitations</strong></h3><p>The Boogu team is candid that 0.1 is an early research release: limited world knowledge for brands, celebrities, and landmarks; instability in strict subject preservation; occasional text typos or layout drift; and artifacts in complex poses, small faces, and limbs.</p><h3 id="links"><strong>Links</strong></h3><ul><li><p>Source: <a target="_blank" rel="ugc" href="https://huggingface.co/Boogu">Hugging Face</a> | <a target="_blank" rel="ugc" href="https://github.com/boogu-project/Boogu-Image">GitHub</a> | <a target="_blank" rel="ugc" href="https://modelscope.cn/organization/Boogu">ModelScope</a></p></li><li><p>Project site: <a target="_blank" rel="ugc" href="https://boogu.org">boogu.org</a></p></li><li><p>Gallery: <a target="_blank" rel="ugc" href="https://boogu-gallery.netlify.app/">boogu-gallery.netlify.app</a></p></li></ul><p><em>Boogu-Image 0.1 is a research project, not an official product release. The Boogu team asks that you not build unauthorized commercial products under the Boogu name. Licensed under Apache 2.0.</em></p>

Boogu-Image Edit-turbo

booguImageEditTurbo_v01_txt.safetensors

booguImageEditTurbo_v01.safetensors

v0.1

<h1 id="qwen3-vl"><span><strong>Qwen3-VL</strong></span></h1><p>Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.</p><p>This generation delivers comprehensive upgrades across the board: superior text understanding &amp; generation, deeper visual perception &amp; reasoning, extended context length, enhanced spatial and video dynamics comprehension, and stronger agent interaction capabilities.</p><p>Available in Dense and MoE architectures that scale from edge to cloud, with Instruct and reasoning‑enhanced Thinking editions for flexible, on‑demand deployment.</p><p><span><strong>Key Enhancements:</strong></span></p><ul><li><p><strong>Visual Agent</strong>: Operates PC/mobile GUIs—recognizes elements, understands functions, invokes tools, completes tasks.</p></li><li><p><strong>Visual Coding Boost</strong>: Generates Draw.io/HTML/CSS/JS from images/videos.</p></li><li><p><strong>Advanced Spatial Perception</strong>: Judges object positions, viewpoints, and occlusions; provides stronger 2D grounding and enables 3D grounding for spatial reasoning and embodied AI.</p></li><li><p><strong>Long Context &amp; Video Understanding</strong>: Native 256K context, expandable to 1M; handles books and hours-long video with full recall and second-level indexing.</p></li><li><p><strong>Enhanced Multimodal Reasoning</strong>: Excels in STEM/Math—causal analysis and logical, evidence-based answers.</p></li><li><p><strong>Upgraded Visual Recognition</strong>: Broader, higher-quality pretraining is able to “recognize everything”—celebrities, anime, products, landmarks, flora/fauna, etc.</p></li><li><p><strong>Expanded OCR</strong>: Supports 32 languages (up from 19); robust in low light, blur, and tilt; better with rare/ancient characters and jargon; improved long-document structure parsing.</p></li><li><p><strong>Text Understanding on par with pure LLMs</strong>: Seamless text–vision fusion for lossless, unified comprehension.</p></li></ul><p><span><strong>Model Architecture Updates:</strong></span></p><p><img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/qwen3vl_arc.jpg" width="80%" /></p><p></p><ol><li><p><strong>Interleaved-MRoPE</strong>: Full‑frequency allocation over time, width, and height via robust positional embeddings, enhancing long‑horizon video reasoning.</p></li><li><p><strong>DeepStack</strong>: Fuses multi‑level ViT features to capture fine‑grained details and sharpen image–text alignment.</p></li><li><p><strong>Text–Timestamp Alignment:</strong> Moves beyond T‑RoPE to precise, timestamp‑grounded event localization for stronger video temporal modeling.</p></li></ol><p><br /></p>

Qwen3-VL

qwen3VL_8b.safetensors

<p>Boogu-Image Edit is the image editing build in the Boogu-Image 0.1 family, a 10B-parameter model that takes a text prompt plus a source image and returns an edited result. It handles object insertion, replacement, and removal, attribute changes, and text editing while keeping the rest of the scene intact.</p>
<p><strong>Originally released by the <a href="https://huggingface.co/Boogu" rel="ugc">Boogu team</a> on Hugging Face.</strong> All credit for the model goes to the Boogu project and its contributors. Civitai is hosting a mirror so creators can run it on-site without a local setup - please head to the <a href="https://github.com/boogu-project/Boogu-Image" rel="ugc">original repo</a> and <a href="https://boogu.org" rel="ugc">project site</a> for weights, updates, and to support the project directly.</p>
<h3><strong>Built by the Boogu project</strong></h3>
<ul>
<li><a href="https://boogu.org" rel="ugc">Boogu project</a> - model training and release</li>
<li><a href="https://huggingface.co/Boogu" rel="ugc">Hugging Face org</a> and <a href="https://modelscope.cn/organization/Boogu" rel="ugc">ModelScope org</a> - weights and variant downloads</li>
<li>Acknowledged upstream influences: Qwen-Image, Z-Image, OmniGen2, FLUX, and DeepSeek</li>
</ul>
<h3><strong>What it does</strong></h3>
<p>Edit is a text-plus-image-to-image model for targeted changes: add or remove objects, swap attributes (color, material, style), and edit rendered text, all driven by a natural-language instruction. It shares the family's bilingual Chinese and English text handling and runs at 25 to 50 steps with CFG in the 2.0 to 5.0 range.</p>
<h3><strong>Versions on Civitai</strong></h3>
<p>We mirror the v0.1 release in three diffusion builds: <code>bf16</code> full precision (~20GB), <code>fp8_scaled</code> (~10GB), and <code>nvfp4</code> (~5.7GB). The Qwen3-VL 8B text encoder and the Flux VAE are bundled so you can assemble a complete editing workflow.</p>
<h3><strong>Related Boogu models</strong></h3>
<p><a href="/models/2714299" rel="ugc">Boogu-Image Base</a> is the text-to-image foundation model. <a href="/models/2714686" rel="ugc">Boogu-Image Turbo</a> is the fast 4-step text-to-image variant.</p>
<h3><strong>Known limitations</strong></h3>
<p>The Boogu team is candid that 0.1 is an early research release: limited world knowledge for brands, celebrities, and landmarks; instability in strict subject preservation; occasional text typos or layout drift; and artifacts in complex poses, small faces, and limbs.</p>
<h3><strong>Links</strong></h3>
<ul>
<li>Source: <a href="https://huggingface.co/Boogu" rel="ugc">Hugging Face</a> | <a href="https://github.com/boogu-project/Boogu-Image" rel="ugc">GitHub</a> | <a href="https://modelscope.cn/organization/Boogu" rel="ugc">ModelScope</a></li>
<li>Project site: <a href="https://boogu.org" rel="ugc">boogu.org</a></li>
<li>Gallery: <a href="https://boogu-gallery.netlify.app/" rel="ugc">boogu-gallery.netlify.app</a></li>
</ul>
<p><em>Boogu-Image 0.1 is a research project, not an official product release. The Boogu team asks that you not build unauthorized commercial products under the Boogu name. Licensed under Apache 2.0.</em></p>

Boogu-Image Edit

booguImageEdit_v01_txt.safetensors

flux1_vae_bf16.safetensors

booguImageEdit_v01.safetensors

<p>Boogu-Image Base is the foundation text-to-image model in the Boogu-Image 0.1 family, a 10B-parameter multimodal system built for photographic realism, dense bilingual text rendering, and broad stylistic range. It is the model to start with when you want maximum diversity and prompt adherence, and the parent that the Turbo and Edit builds derive from.</p>
<p><strong>Originally released by the <a href="https://huggingface.co/Boogu" rel="ugc">Boogu team</a> on Hugging Face.</strong> All credit for the model goes to the Boogu project and its contributors. Civitai is hosting a mirror so creators can run it on-site without a local setup - please head to the <a href="https://github.com/boogu-project/Boogu-Image" rel="ugc">original repo</a> and <a href="https://boogu.org" rel="ugc">project site</a> for weights, updates, and to support the project directly.</p>
<h3><strong>Built by the Boogu project</strong></h3>
<ul>
<li><a href="https://boogu.org" rel="ugc">Boogu project</a> - model training and release</li>
<li><a href="https://huggingface.co/Boogu" rel="ugc">Hugging Face org</a> and <a href="https://modelscope.cn/organization/Boogu" rel="ugc">ModelScope org</a> - weights and variant downloads</li>
<li>Acknowledged upstream influences: Qwen-Image, Z-Image, OmniGen2, FLUX, and DeepSeek</li>
</ul>
<h3><strong>What it does</strong></h3>
<p>Base targets photography-grade output with natural lighting and composition, bilingual Chinese and English text rendering across complex layouts, stylization across many artistic genres, and poster and product design. It runs at 25 to 50 steps with CFG in the 2.0 to 5.0 range.</p>
<h3><strong>Versions on Civitai</strong></h3>
<p>We mirror the v0.1 release in three diffusion builds: <code>bf16</code> full precision (~20GB), <code>fp8_scaled</code> (~10GB), and <code>nvfp4</code> (~5.7GB). The required Qwen3-VL 8B text encoder and the Flux VAE are bundled so you can assemble a complete workflow.</p>
<h3><strong>Related Boogu models</strong></h3>
<p><a href="/models/2714686" rel="ugc">Boogu-Image Turbo</a> is the distilled 4-step variant for fast generation. <a href="/models/2714541" rel="ugc">Boogu-Image Edit</a> adds image-to-image editing.</p>
<h3><strong>Known limitations</strong></h3>
<p>The Boogu team is candid that 0.1 is an early research release: limited world knowledge for brands, celebrities, and landmarks; instability in strict subject preservation; occasional text typos or layout drift; and artifacts in complex poses, small faces, and limbs.</p>
<h3><strong>Links</strong></h3>
<ul>
<li>Source: <a href="https://huggingface.co/Boogu" rel="ugc">Hugging Face</a> | <a href="https://github.com/boogu-project/Boogu-Image" rel="ugc">GitHub</a> | <a href="https://modelscope.cn/organization/Boogu" rel="ugc">ModelScope</a></li>
<li>Project site: <a href="https://boogu.org" rel="ugc">boogu.org</a></li>
<li>Gallery: <a href="https://boogu-gallery.netlify.app/" rel="ugc">boogu-gallery.netlify.app</a></li>
</ul>
<p><em>Boogu-Image 0.1 is a research project, not an official product release. The Boogu team asks that you not build unauthorized commercial products under the Boogu name. Licensed under Apache 2.0.</em></p>

Boogu-Image Base

booguImageBase_v01.safetensors

booguImageBase_v01_txt.safetensors

<p>Boogu-Image Turbo is the speed-distilled text-to-image build in the Boogu-Image 0.1 family. It carries the same 10B-parameter backbone as <a href="/models/2714299" rel="ugc">Boogu-Image Base</a> but is tuned to produce photorealistic results in just 4 steps with CFG 0, trading some of Base's diversity for a large jump in generation speed.</p>
<p><strong>Originally released by the <a href="https://huggingface.co/Boogu" rel="ugc">Boogu team</a> on Hugging Face.</strong> All credit for the model goes to the Boogu project and its contributors. Civitai is hosting a mirror so creators can run it on-site without a local setup - please head to the <a href="https://github.com/boogu-project/Boogu-Image" rel="ugc">original repo</a> and <a href="https://boogu.org" rel="ugc">project site</a> for weights, updates, and to support the project directly.</p>
<h3><strong>Built by the Boogu project</strong></h3>
<ul>
<li><a href="https://boogu.org" rel="ugc">Boogu project</a> - model training and release</li>
<li><a href="https://huggingface.co/Boogu" rel="ugc">Hugging Face org</a> and <a href="https://modelscope.cn/organization/Boogu" rel="ugc">ModelScope org</a> - weights and variant downloads</li>
<li>Acknowledged upstream influences: Qwen-Image, Z-Image, OmniGen2, FLUX, and DeepSeek</li>
</ul>
<h3><strong>What it does</strong></h3>
<p>Turbo is the build to reach for when throughput matters: fast iteration, batch work, and interactive generation. It keeps the family's strengths in photography, bilingual Chinese and English text rendering, stylization, and poster design, at a fraction of the step count. Recommended starting point is 4 steps, CFG 0.</p>
<h3><strong>Versions on Civitai</strong></h3>
<p>We mirror the v0.1 release in three diffusion builds: <code>bf16</code> full precision (~20GB), <code>fp8_scaled</code> (~10GB), and <code>nvfp4</code> (~5.7GB). The required Qwen3-VL 8B text encoder and the Flux VAE are bundled. Note: this is a full distilled checkpoint, not a LoRA.</p>
<h3><strong>Related Boogu models</strong></h3>
<p><a href="/models/2714299" rel="ugc">Boogu-Image Base</a> is the full-quality foundation model at 25 to 50 steps. <a href="/models/2714541" rel="ugc">Boogu-Image Edit</a> adds image-to-image editing.</p>
<h3><strong>Known limitations</strong></h3>
<p>The Boogu team is candid that 0.1 is an early research release: limited world knowledge for brands, celebrities, and landmarks; instability in strict subject preservation; occasional text typos or layout drift; and artifacts in complex poses, small faces, and limbs.</p>
<h3><strong>Links</strong></h3>
<ul>
<li>Source: <a href="https://huggingface.co/Boogu" rel="ugc">Hugging Face</a> | <a href="https://github.com/boogu-project/Boogu-Image" rel="ugc">GitHub</a> | <a href="https://modelscope.cn/organization/Boogu" rel="ugc">ModelScope</a></li>
<li>Project site: <a href="https://boogu.org" rel="ugc">boogu.org</a></li>
<li>Gallery: <a href="https://boogu-gallery.netlify.app/" rel="ugc">boogu-gallery.netlify.app</a></li>
</ul>
<p><em>Boogu-Image 0.1 is a research project, not an official product release. The Boogu team asks that you not build unauthorized commercial products under the Boogu name. Licensed under Apache 2.0.</em></p>

Files with SHA256: 4ba424cf62e51392e4d1a39933e803706f4e823c1065f36aaf149c6453f66bcd