Many advanced multimodal AI platforms can already understand complex instructions, generate realistic images, edit existing visuals, and handle text inside images. However, these capabilities often come from highly integrated systems that combine powerful models, massive datasets, and expensive training pipelines. The problem is not only about building a larger model. It is about improving the entire system like data quality, training strategy, and the way generation and editing capabilities are combined.
Boogu Image released by the Boogu Team, a 10-billion-parameter open-source image generation and editing model family designed to provide practical, high-quality multimodal capabilities.
 |
| Boogu Image model showcase |
Instead of focusing on only one task, the model family includes multiple variants:
- Base for high-quality image generation
-Turbo for faster image creation
-Edit for image modification tasks.
The goal is to provide creators, developers, and researchers with a unified system that can generate, understand, and edit images without relying on closed-source solutions.
Compared with some existing open-source models, Boogu-Image-0.1 was trained with roughly one order of magnitude less data, yet it still achieves competitive results. For the best output quality, the team recommends running Boogu-Image-0.1-Base at 2K resolution, especially when layout accuracy and character consistency are important.
Installation
1. Make sure you have done ComfyUI installation. Older user should update ComfyUI from the Manager.
2. Download Boogu image (base/edit/turbo) model from Hugging face repository. Choose as per your requirements and system resources:
(a) boogu_image_base_bf16.safetensors - BF16 base (for best img generation)
(b) boogu_image_base_fp8_scaled.safetensors - FP8 base (for optimized img generation)
(c) boogu_image_base_nvfp4.safetensors - NVFP4 base (for fast img generation with quality compromised)
(d) boogu_image_edit_bf16.safetensors - BF16 edit (for best editing with generation)
(e) boogu_image_edit_fp8_scaled.safetensors -FP8 edit (for optimized editing with generation)
(f) boogu_image_edit_nvfp4.safetensors -NVFP4 edit (for fast editing with generation with quality degradation)
(g) boogu_image_turbo_bf16.safetensors - BF16 Turbo (for best generation)
(h) boogu_image_turbo_fp8_scaled.safetensors - FP8 scaled Turbo (for optimized and fast generation)
(i) boogu_image_turbo_nvfp4.safetensors - NVFP4 Turbo (for fastest generation with quality compromised)
Save them into ComfyUI/models/diffusion_models folder.
3. Download text encoder (qwen3vl_8b_fp8_scaled.safetensors) and save this into ComfyUI/models/text_encoders folder. You do need to download it if already have.
4. Download Vae (flux1_vae_bf16.safetensors) and save this into ComfyUI/models/vae folder. You do need to download it if already have.
5. Download lora (boogu_image_turbo_lora_rank_128_bf16.safetensors) and save this into ComfyUI/models/loras folder. You can use this as optional model. This lora model can be used to apply the turbo distillation feature on the base/edit model. For edit model, use it as half the strength with little more steps.
6. Refresh and restart comfyui to take effect.
Workflow
1. Download workflows (edit/base/turbo) variants from our hugging face repository:
(a) Boogu-Image_Base.json
(b) Boogu-Image_Edit.json
(c) Boogu-Image_Turbo.json
2. Drag and drop the workflow into ComfyUI.
3. Load the models into its relevant nodes.
4. Set configuration for different variants-
(a)Boogu Image Base settings-
CFG-4.0
Steps-30 to 50
Resolution-1k to 2k (maximum for best generation)
(b)Boogu Image Turbo settings-
CFG-1.0
Steps-4
Scheduler-sgm_uniform
Resolution-1k to 2k (max)
(c)Boogu Image Edit settings-
CFG-3.5
Steps-25
Scheduler-simple
5. Put prompts into positive and negative prompt box. The model supports both the languages-English and Chinese. Make sure you use long detailed prompts to get the best out of it. Normal prompting will generate bad results.
6. Hit run to start the generation.
Some of the generation we did:
 |
| Image Generated using Boogu Image Turbo |
Prompt:
The image is an advertisement for FUME perfume, featuring a soft portrait bathed in the cool red light of a window. In the upper left, a large white serif headline reads "FUME -FOR YOUNG SKIN". The main subject is a american young woman resting on a light-colored counter; her long hair is slightly damp, and she gazes at the camera with her cheek pressed against her arm, set against a backdrop of vertical window frames and soft-focus light and shadow. In the left foreground stands a tube of red face wash—labeled "FUME," "TOTAL CARE FACE," "PARIS," "MINT," and "120g / 5.23 oz"—with a small dollop of white paste resting on the counter beside it. Along the left edge, the English phrase "FUME, FOR THE MOMENTS THAT MATTER" appears vertically. Large text at the bottom reads "Define Your Moment," followed by the slogan "Define every moment of your radiance," and finally "# Everyday Confidence" and "FROM PARIS." The overall aesthetic is clean, serene, and sophisticated, characterized by cool tones.
 | |
|
Prompt:
A highly structured, portrait-oriented infographic poster titled '12 MAKEUP LOOKS FOR SPRING'. The background is a crisp, soft pastel pink. Top header section: On the left, a large stylish number '12' followed by the title 'MAKEUP LOOKS / FOR SPRING' in an elegant serif font, with a smaller subtitle 'Discover Your Vibe' below it. On the top right, small text reads 'FRESH • GLOWING • VIBRANT'. The main body consists of a 4x3 grid of rectangular portrait panels. Each panel features a high-quality, soft-focus portrait of the same young Caucasian woman with blonde hair, wearing a simple silk camisole, against a luminous studio background. Each panel demonstrates a different makeup style on her, and has a number (01 to 12) in the top-left corner, and a semi-transparent text box overlay at the bottom with the look name in bold, a short description, and occasion tags. Row 1: panel 01 'PEACH GLOW' (description 'Warm & radiant', tags 'BRUNCH / PICNIC') showing her smiling with peachy blush and glossy peach lips; panel 02 'DEWY N*DE' (description 'Fresh & minimal', tags 'GYM / ERRANDS') showing glass skin, brushed-up brows, and clear lip balm; panel 03 'ROSE PETAL' (description 'Soft & romantic', tags 'DATE / WEDDING') showing her with rosy pink tones on her cheeks and a soft blurred pink lip. Row 2: panel 04 'CORAL POP' (description 'Bright & fun', tags 'PARTY / FESTIVAL') looking playful with vibrant coral lips and matching eyeshadow; panel 05 'SOFT GLAM' (description 'Polished & chic', tags 'WORK / MEETING') with neutral taupe eyeshadow and a nude matte lip; panel 06 'BERRY STAIN' (description 'Deep & alluring', tags 'NIGHT OUT') showing a sultry gaze with a bitten berry lip tint. Row 3: panel 07 'SUN-KISSED' (description 'Bronzed & healthy', tags 'BEACH / VACATION') with warm bronzer across her nose and faux freckles; panel 08 'LILAC HAZE' (description 'Cool & trendy', tags 'FASHION / EVENT') with soft purple lilac eyeshadow and pale pink lips; panel 09 'BOLD WING' (description 'Sharp & edgy', tags 'CLUB / CONCERT') with dramatic black winged eyeliner and a bare lip. Row 4: panel 10 'GLOSSY CHERRY' (description 'Juicy & sweet', tags 'DATE / DAILY') with high-shine red cherry lip gloss; panel 11 'MATTE TAUPE' (description 'Sophisticated & modern', tags 'OFFICE / DINNER') with cool matte taupe eyeshadow and a muted brown lip; panel 12 'GOLDEN HOUR' (description 'Shimmering & warm', tags 'GALA / RED CARPET') showing intense gold highlighter on the cheekbones, inner eye corners, and cupid's bow.
The biggest takeaway is that progress does not always come from simply making models bigger. Better data, smarter training approaches, and stronger integration between understanding and generation can create significant improvements.
A unified image model that can generate visuals, edit images, and handle complex text layouts moves AI closer to becoming a complete creative assistant rather than just a generation tool.
-BF16/FP8/NVFP4 ){kind=link}