nanobanana
Skill11 starsChanged 6 months agoISCView on GitHub
Craft high-precision prompts and edit instructions for Nano Banana image workflows, especially when using the local nanobanana MCP tools for generation, editing, character consistency, or multi-image fusion. Use when the task needs structured prompts, reference-role assignment, layout-heavy image specs, typography-heavy images, iterative edit-first refinement, or reliable model/aspect/output-path choices.
12 other projects use a version of this.
Adopted by
Most recently taken up by
- jpcaparas/skillstheir own version, first seen Nov 15, 2026
- amigoscode/skillstheir own version, first seen Oct 13, 2026
- team-attention/stanford-cs146s-krtheir own version, first seen Oct 8, 2026
- Xipher-Labs/walter-ostheir own version, first seen May 21, 2026
- link2004/skillstheir own version, first seen May 9, 2026
- grahac/claude_skillstheir own version, first seen Apr 18, 2026
Other versions12 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 12
- ReScienceLab/opc-skills★ 1.9ksame name
- team-attention/stanford-cs146s-kr★ 294same name
- jpcaparas/skills★ 57same name
- maragudk/skills★ 49same name
- link2004/skills★ 17same name
- Xipher-Labs/walter-os★ 8same name
- ImScottDev/brain-scott★ 4same name
- grahac/claude_skills★ 2same name
- amigoscode/skills★ 1same name
- dortort/skills★ 1same name
2 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- kling-3-0-deep-diveSkill
- media-projectSkill
- phoneticizeSkill
- scene-inventory-extractorSkill
- seedance-2-deep-diveSkill
- shot-specifierSkill
- video-generatorSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
15 rules. 0 are common, 11 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 10 edited versions we read keep these.
11
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use this skill to turn a vague image request into a prompt that Nano Banana can execute reliably, or to refine an existing image/edit …
- Distinctive: line 21
Use a short natural-language prompt for simple scenes.
- Distinctive: line 22
Use a structured block for multi-part layouts, infographics, or scene logic.
- Distinctive: line 23
Use JSON-like structure when the user needs many simultaneous constraints or multiple reference roles.
- Distinctive: line 35
Prefer positive, explicit constraints.
- Distinctive: line 43
If the first output is close, preserve the successful parts and request only the delta.
- Very specific: line 53
Every image-generation or image-editing MCP call must explicitly pass model: gemini-3-pro-image-preview.
- Very specific: line 59
If gemini-3-pro-image-preview is unavailable, rejected by the tool, or cannot accept the required reference images, character-consistency …
- Very specific: line 66
Use gemini-3-pro-image-preview for all Nano Banana image workflows, including generation, editing, character consistency, and multi-image …
- Very specific: line 68
Do not choose a speed, aspect-ratio, or fast-iteration fallback model.
- Distinctive: line 80
Do not overstuff the prompt with decorative synonyms when concrete constraints will do.
- Distinctive: line 84
When an example from the source material is unsafe, irrelevant, or too verbose for direct reuse, extract the pattern and rewrite it into a …
- Distinctive: line 88
If the user asks for an image, provide a prompt plus suggested model, aspect ratio, and any reference-role mapping.
- Distinctive: line 89
If the user asks for an edit, provide the preservation constraints first, then the requested changes.
- Distinctive: line 90
If the user asks for multiple options, vary composition and lighting before varying everything else.
“Common” is measured: kept by at least 6 in 10 of the 10 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Nanobanana
Use this skill to turn a vague image request into a prompt that Nano Banana can execute reliably, or to refine an existing image/edit request without restarting from scratch.
Workflow
- Identify the operation first.
generate: create a new image from text and optional references.edit: modify an existing image while preserving specified elements.character_consistency: keep one character stable across scenes.multi_image_fusion: combine several references into one coherent result.
- Choose the lowest-complexity prompt shape that fits the ask.
- Use a short natural-language prompt for simple scenes.
- Use a structured block for multi-part layouts, infographics, or scene logic.
- Use JSON-like structure when the user needs many simultaneous constraints or multiple reference roles.
- Build the prompt in this order.
- Subject: who or what must appear.
- Action/state: what is happening.
- Setting: where it is happening.
- Composition: framing, camera angle, spatial layout.
- Lighting/mood: time of day, light direction, contrast, atmosphere.
- Style/materiality: photoreal, editorial, infographic, matte acrylic, film grain, etc.
- Constraint layer: exact text, counts, positions, identity preservation, what must remain unchanged.
- Prefer positive, explicit constraints.
- Say what should be present and how it should behave.
- For edits, explicitly state what must remain unchanged.
- For text rendering, quote exact text and specify placement plus typographic character.
- Use edit-first iteration.
- If the first output is close, preserve the successful parts and request only the delta.
- Example:
Keep composition, subject identity, and wardrobe identical; shift lighting to golden hour and replace background with a foggy bridge.
Local MCP Rules
mcp__nanobanana__generate_imageis for new images.mcp__nanobanana__edit_imageis for targeted changes to an existing image.mcp__nanobanana__character_consistencyis for repeated scenes with one character reference.mcp__nanobanana__multi_image_fusionis for combining several references.output_pathmust stay inside the tool's allowed repo-local output area. In practice, use a simple filename or relative path; do not pass an absolute path outsideimage_out.- Every image-generation or image-editing MCP call must explicitly pass
model: gemini-3-pro-image-preview. - Pass image inputs through
referenceImagePathswith the smallest complete set of references for the operation. Do not throw every available reference into a general pool; assign only the location, prop, style, and character anchors that constrain the current frame. - If
gemini-3-pro-image-previewis unavailable, rejected by the tool, or cannot accept the required reference images, character-consistency images, or multi-image-fusion inputs for the requested operation, stop the workflow and report the blocker. Do not fall back to another image model.
Model Selection
- Use
gemini-3-pro-image-previewfor all Nano Banana image workflows, including generation, editing, character consistency, and multi-image fusion. - Do not choose a speed, aspect-ratio, or fast-iteration fallback model. If the required
image operation cannot run on
gemini-3-pro-image-preview, stop and surface the exact limitation.
Prompting Patterns
- For simple generation, use the formula in references/frameworks.md.
- For structured scene blueprints, multi-reference prompts, and edit instructions, use references/frameworks.md.
- For reusable prompt skeletons across posters, product shots, infographics, floor plans, storyboards, and character scenes, use references/examples.md.
Guardrails
- Do not overstuff the prompt with decorative synonyms when concrete constraints will do.
- For exact counts, rows, layouts, labels, or room sizes, state them numerically and spatially.
- For reference images, assign each image a job such as identity, pose, style, lighting, or environment.
- For text-heavy images, keep each required text string short unless the user explicitly needs a dense poster or infographic.
- When an example from the source material is unsafe, irrelevant, or too verbose for direct reuse, extract the pattern and rewrite it into a safe, shorter template rather than copying it.
Deliverable Style
- If the user asks for an image, provide a prompt plus suggested model, aspect ratio, and any reference-role mapping.
- If the user asks for an edit, provide the preservation constraints first, then the requested changes.
- If the user asks for multiple options, vary composition and lighting before varying everything else.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

