1/8 参数，跑赢 80B 大模型：Boogu-Image 是黑马还是鸡肋？
Ranking
Overall
68
Content
75
Popularity
N/A
No observed public metrics; popularity remains neutral/archived.
Merged summary
TL;DR - Open-source Boogu-Image-0.1 outperforms larger image models on Qwen-Image-Bench with 10B parameters, but hands-on testing finds it not yet reliable for production delivery. Its strongest value is as a low-cost, frequently invoked image module in agent workflows.
- It scores 53.58 on Qwen-Image-Bench, ahead of 20B Qwen-Image-2512 and 80B Hunyuan-Image-3.0.
- Base, Turbo, Edit, and FP8 variants target dense text rendering, 3–4-step generation, localized editing, and lower-memory deployment.
- Tests showed strong photorealism, composition, and simple edits, but persistent weaknesses in text accuracy, physical consistency, complex edits, and cinematic polish.
- The 10B/FP8 combination could enable private deployment and high-volume agent sampling, where latency and cost matter more than best-in-class single-image quality.
Sources (1)
1/8 参数，跑赢 80B 大模型：Boogu-Image 是黑马还是鸡肋？
Public signals
N/A
TL;DR - Open-source Boogu-Image-0.1 outperforms larger image models on Qwen-Image-Bench with 10B parameters, but hands-on testing finds it not yet reliable for production delivery. Its strongest value is as a low-cost, frequently invoked image module in agent workflows.
- It scores 53.58 on Qwen-Image-Bench, ahead of 20B Qwen-Image-2512 and 80B Hunyuan-Image-3.0.
- Base, Turbo, Edit, and FP8 variants target dense text rendering, 3–4-step generation, localized editing, and lower-memory deployment.
- Tests showed strong photorealism, composition, and simple edits, but persistent weaknesses in text accuracy, physical consistency, complex edits, and cinematic polish.
- The 10B/FP8 combination could enable private deployment and high-volume agent sampling, where latency and cost matter more than best-in-class single-image quality.