wzc520pyfm/gen-ai-pose-image

Lanpose is a browser studio — also packaged as a DeepSeek Harness plugin — for generating images of a single character model in chosen poses and outfits. It pulls pose line art and costume references from two external public galleries and sends them, together with a model reference, to an image-editing API. It fits people who want repeatable, descriptor-driven character posing rather than free-form prompting.

What it does

The default model is Dafeiyu, a community whale-girl character with blue wavy hair, whale ears and a whale tail in a navy-and-white maid outfit; its reference sheet comes from the separate deepseek-whale-girl project README, deepseek-whale-girl. You pick one or more poses from the ai-pose-gallery-zh set of 288 pose groups / 1,439 line drawings, and optionally one costume from the COSTUME-DB-01 collection of 1,407 outfits, which contain no person ai-pose-gallery-zh, COSTUME-DB-01. You can also import up to three images of another model to replace the face and clothes.

How it works

Pose sketches are not stored in the repository; the page reads them on demand from the public image host, and similarly loads costume thumbnails, keeping only IDs and keywords locally README. The pose image supplies body posture only, and blank-body areas in the sketch are not meant to be rendered by default. When a costume is selected it temporarily overrides the "keep outfit" option, while the character's face, hairstyle, whale ears and tail are kept.

Image generation needs a key for one of three providers: Seedream via Volcengine Ark (default model doubao-seedream-5-0-flash-260915), Gemini (default gemini-2.5-flash-image), or an OpenAI-compatible endpoint. The OpenAI path first tries images/edits and, if editing is unsupported, falls back to plain text-to-image while stating in the result that reference images were not submitted. Without any key, the studio still writes out a ready-to-copy prompt for each image — for example to pass to dsh-image-gen — with references ordered character, person-free costume, pose sketch README.

Adult-category poses and costumes (lingerie, swimwear, sexy categories) require explicitly turning on the adult switch, and are excluded from search results unless requested.

Setup paths and requirements

• Local studio: needs Node.js 22.19 or newer, then npm install and npm run dev, served at a local address; the layout adapts to narrow screens and phones README, studio screenshots.
• DeepSeek Harness plugin: install into an existing DeepSeek Harness profile with dsh plugin --profile web add gen-ai-pose-image; a restart of that profile adds a "Lanpose / Studio" sidebar entry and tools search_poses, search_costumes and generate_pose_images Harness screenshot, README.
• Key handling: keys entered in the browser stay in that browser; alternatively a server can hold them in environment variables. In the Harness context, setting the key lets tools generate and attach images directly; without it they return prompts.

Licensing and scope

The program is MIT-licensed, but that covers only the code. The pose line art and original data remain the property of their author, and their upstream notice states the images may not be used commercially; costume images come from the same author; the Dafeiyu character sheet is used under the deepseek-whale-girl MIT license and is a community figure, not an official DeepSeek representation LICENSE, README. The repository lists no independent quality evaluation; the README's examples are author-produced results.