Developer Builds Stateful Image-Editing Skill for Claude Code Using Google Gemini API

·1 views

A developer has released an open-source project called nb2lite-skill-claude that integrates Google's Gemini image model into Claude Code as a stateful image-editing skill. The tool wraps the gemini-3.1-flash-lite-image model in a FastMCP server, allowing users to generate and iteratively refine images directly within a Claude Code session. Unlike typical stateless image workflows, it leverages Google's Interactions API, which preserves visual context server-side across multiple editing turns using a chained interaction ID system. Users can issue short follow-up prompts like 'add a lantern in its paw' without re-describing the full scene, as the model retains character, lighting, and composition from prior turns. The project ships as both a Model Context Protocol server and a Claude Code skill, making it compatible with any MCP-capable AI client.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How a Simple Image Budget Can Keep Your Website Fast Long-Term

Web teams often treat image optimization as a last-minute fix, compressing a few files when a page feels slow before new unoptimized images quickly undo the work. A more sustainable approach involves creating an 'image budget' — a defined set of rules covering file size, dimensions, format, and visual quality that applies before publication. The process begins by classifying images by purpose, since hero photos, screenshots, icons, and diagrams each have different compression tolerances and quality requirements. Teams are advised to set a total page transfer allowance first, then divide that budget across individual assets rather than applying a single compression setting to every file. Writing these limits in plain, checkable language gives writers, designers, and developers a shared standard, turning optimization from a reactive rescue task into a routine publishing decision.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Five Ways to Access European Bank Data in 2026 and What Each Costs

Fintech developers building apps in Europe can access bank transaction data through five main routes under the PSD2 directive, each carrying different costs and complexity. Direct AISP registration offers the most control but requires €4,000–€16,000 or more annually in certificates and legal fees, plus significant engineering time per bank. Aggregator platforms like Plaid, TrueLayer, and Yapily simplify integration to days but charge $0.50–$2.00 per connected account monthly, which scales steeply with user growth. Nordigen, once known for a generous free tier, has shifted pricing focus toward enterprise clients following its acquisition by GoCardless. Middleware API providers represent a fifth option, abstracting away certificate requirements entirely and offering simple API-key access for developers who want to avoid regulatory overhead.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

TypeScript 5.5 Introduces Inferred Type Predicates to Simplify Type Narrowing

TypeScript 5.5 has introduced inferred type predicates, a feature that allows the compiler to automatically determine a value's type based on developer-defined conditions. Previously, developers had to write explicit type definitions or manual checks, making code more verbose and harder to maintain. With type predicate functions, TypeScript can now narrow types during operations like array filtering without requiring redundant annotations. The feature supports complex use cases, including generics, conditional types, and nested conditions, broadening its practical utility. This update is aimed at improving the overall developer experience by making TypeScript code more concise and easier to manage.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developers Build Hybrid AI System to Predict Athlete and Coder Burnout Early

A technical guide published on DEV Community outlines how to build a burnout early warning system by combining Facebook Prophet and PyTorch Transformers to forecast physiological fatigue. The system ingests Heart Rate Variability data from the Oura Ring wearable device via its API, using Prophet to detect macro-level trends such as weekly workout cycles and monthly stress patterns. A PyTorch Transformer model simultaneously analyzes subtle, non-linear HRV drops over a 14-day window to identify micro-signals of nervous system stress. The two models feed into a feature fusion layer that outputs a burnout risk percentage, prompting actionable recommendations like rest or active recovery. The approach aims to shift users from reactive readiness scores to predictive fatigue alerts before physical exhaustion sets in.

0 comments Read more at DEV Community