ZHO’s Nano Banana Creation Collection: A 3.7k-Star AI Art Prompt Treasure Trove

Anyone who plays with AI image generation knows that prompt quality directly determines output quality. The same model can produce either photorealistic masterpieces or blurry messes — the difference is all in prompt engineering.

Recently stumbled upon a gem on GitHub: ZHO-ZHO-ZHO/ZHO-nano-banana-Creation. 3.7k stars, a pure Chinese community project collecting creative workflows and high-quality prompts for Doubao’s Nano Banana Pro model. After going through it, my output quality noticeably improved.

What problem does it solve

Nano Banana Pro (Doubao’s image generation model) is quite capable, but official docs don’t say much about prompt techniques. Many users get the model but have no idea how to write prompts that actually unlock its potential.

This project addresses exactly that: curating community-verified high-quality prompts and techniques so beginners can get started quickly and veterans can discover new tricks. No need to figure everything out from scratch — stand on the shoulders of giants.

Core content

Photorealistic portraits The collection has an entire category of prompts for generating realistic portrait photos:

• Different lighting conditions (natural light, studio, night, backlight)
• Various focal length simulations (35mm, 50mm, 85mm, telephoto compression)
• Film style emulation (Kodak Portra, Fuji Velvia, B&W grain)
• Professional photography terminology in prompts (bokeh, depth of field, golden hour, blue hour)

I tried the “85mm f/1.4 natural light portrait” prompt and the depth-of-field effect and skin tone rendering were genuinely impressive — could pass for a real photo at first glance.

Concept design and illustration Beyond realistic photos, the collection covers various illustration and concept design styles:

• Cyberpunk cityscapes
• Fantasy creature design
• Chinese traditional/modern fusion (国潮) art
• Flat commercial illustration
• 3D rendering style
• Pixel art

Each style comes with prompt templates and parameter suggestions. Swap a few keywords for instant variations.

Advanced techniques The author doesn’t just dump prompts — shares model-specific tricks too:

• How to use negative prompts to control unwanted elements
• Combining multiple style keywords for hybrid effects
• How resolution parameters affect detail
• Batch generation parameter optimization

Continuous updates The project description says “continuously updating” and it actually is. Author ZHO regularly adds new creative workflows and adapts to model version updates. I’ve seen two content updates in the two weeks I’ve been following.

Real-world usage

Method 1: Direct template use Find a prompt template close to your needs, swap out the subject keywords, generate. Fastest approach for time-pressed scenarios.

Method 2: Deconstruct and learn Instead of copying blindly, analyze the structure of each high-quality prompt — where the subject description goes, how style keywords are arranged, the ordering of lighting and composition terms. Understanding the structure lets you write your own quality prompts.

Method 3: Hybrid innovation Mix elements from different prompt categories. For example, combine portrait lighting descriptions with concept design fantasy elements to create a unique “fantasy portrait” style.

The good and the bad

What I loved:

• Chinese project with Chinese prompts and explanations — friendly for domestic users
• Prompt quality is genuinely high, many are the author’s carefully tuned results
• Comprehensive style coverage from realistic to abstract
• Continuous updates keeping pace with model iterations
• Completely open source and free, zero restrictions
• Decent community vibe, people share their variations in issues

What frustrated me:

• No systematic category index, finding specific styles requires scrolling
• Some prompts are version-specific, newer model versions may produce different results
• Some advanced prompts need specific parameter settings with insufficient documentation
• Output depends on model updates, prompts may become less effective after model changes
• No English version, international users can’t read it
• No search function, browsing efficiency drops as content grows

Compared to other prompt resources

Resource	Characteristics	Best for
ZHO’s collection	Nano Banana Pro focused, Chinese, updated regularly	Domestic users, Doubao ecosystem
PromptHero	Multi-model, community-driven, with image previews	Creators needing inspiration
Civitai	Mainly Stable Diffusion, rich model collection	SD users
Midjourney official docs	Authoritative, systematic, but MJ-only	Midjourney users

My take: if you primarily use Doubao’s Nano Banana Pro model, this collection is a must-bookmark resource. The prompt quality and specificity are noticeably better than generic prompt sharing sites.

Bottom line

3.7k stars is solid for a pure Chinese prompt collection. ZHO’s project delivers the most value by lowering the learning curve for Nano Banana Pro — you don’t need to spend dozens of hours figuring out prompt writing, just reference community-verified high-quality examples.

That said, be realistic: prompt engineering is a moving target. Model updates and algorithm changes affect prompt effectiveness. This collection provides an excellent starting point, but true mastery requires understanding the underlying logic, not just copy-pasting.

For content creators, designers, or AI art enthusiasts who frequently need high-quality image generation, this collection is worth bookmarking. Even if you don’t read it all, knowing it exists means you’ll have a go-to resource when you hit a creative wall.

---

About the Author

Liudingyu is a full-stack developer and heavy GitHub user. With 900+ starred repos over the past 3 years, this site only covers tools I’ve actually used or deeply researched.

📧 Found a great tool to recommend? Email [email protected]