Neko3000/deepseek-whalechan

What it solves

DeepSeek Whale-chan addresses the problem of keeping an AI-generated anime character visually consistent across many images. Without a strict specification, image models tend to drift in hair color, costume details, and body proportions. This project provides a formal character design blueprint—covering hair, whale-fin ears, cowlick, tail, and maid uniform—plus five quantified head-to-body ratios (from 4.0-head standard to 2.1-head super-deformed) to enforce uniformity. It also supplies a comedy engine that preserves humor and personality when turning technical chats, error logs, or user rants into multi-panel comics.

How it works

The project defines a complete character profile, including personality traits, classic quotes, and visual assets (reference sheets for poses, expressions, fins, tail motions, and costumes). It provides a set of Agent Skills that work with external image generation providers (Codex ImageGen, OpenAI, Nano Banana, Seedream). For portraits, the whalechan-image-character skill runs a pipeline: multiple proposal selection, scope confirmation, then generation and quality assurance using deterministic checks (skeletal joint fitting, native-res visual QA). For comics, the whalechan-image-comic skill freezes a set of ideas, uses a sub-agent to rank them, and generates a comic per selected idea (default 5), with a single user confirmation. All output follows comedy directions—reversal, exposure, escalation, recognition, character gag—and exaggeration tiers (A: manga impact, B: comic outburst, C: restrained deadpan).

Who it's for

Creators, fan-artists, and developers who want to mass-produce consistent illustrations, posters, stickers, or comedy comics featuring the Whale-chan mascot. It's also for anyone exploring AI-driven character consistency and automated visual content pipelines, especially those who value deterministic control over output style and proportions.

Highlights

• Five strictly defined character forms (standard, compact, semi-chibi, chibi, super-deformed) with precise head-to-body ratio acceptance ranges.
• A comprehensive visual asset library: reference sheets, widescreen banners, portrait posters, social square posts, and showcase galleries.
• A character mindset engine that injects the character's personality and comedy style into generated works, with five predefined joke directions and three exaggeration tiers.
• Two out-of-the-box Agent Skills (character portraits and multi-panel comics) that integrate with multiple image providers and include built-in quality gate.
• Community-driven and non-official, with content licensed under CC-BY-NC-SA 4.0 for character IP and fanwork copyright attribution.