Preface

People working on short dramas, comic dramas, and animation all encounter a common problem: the same character looks different in every episode, every panel, and every image. Using general text-to-image models to draw image by image means facial consistency relies on luck. Using reference image workflows like IP-Adapter requires a stable baseline image with clear features, free of extra emotions and backgrounds. Character LoRA training is similar; if the training materials contain too many varied expressions and angles, the results can become erratic.

This Skill’s approach is: to generate a single 1:1 canvas with a 4x4 grid of reference images at once. It draws close-ups of facial features, multi-angle face shapes, and front/side/back three-view drawings into a single image, while forcibly locking the facial bone structure, proportions, and age. Characters are uniformly presented in light gray and without expression. The following section introduces this Skill.

What is this

This is a Skill for generating character consistency reference images, with the coordinates @org-bzwfrdvh/aipdd-char-view. It is published by AIPDD, with a source type of enterprise, a classification of design-media, version 1.0.3, and is based on the general SKILL.md format.

The problem it solves is: providing a baseline image with locked features and a unified style for subsequent character consistency generation, IP-Adapter reference image workflows, character LoRA training, and modeling references, rather than drawing a new face from random noise every time.

Core Features

  1. A single 1:1 canvas with a 4x4 grid layout, simultaneously containing close-ups of facial features, multi-angle face shapes, and front/side/back three-view drawings.
  2. Forcibly locks facial bone structure, proportions, and age to reduce drift of the same character across different generation batches.
  3. Characters are presented in light gray and without expression, reducing interference from background, clothing, and expressions on subsequent reference image workflows.
  4. The output can be used for character consistency generation in short dramas, comic dramas, and animation, and can also be used for character LoRA training materials and modeling references.

Note that what is locked here is consistency at the level of facial structure, not fixing a specific real-person photo.

Model Base and Image Generation Priority

The model base is the sub-skill newapi-media-1.0.19, located in model-invoker/. The default model gateway is AIPDD, covering full modalities including text, vision, image generation, image editing, video, speech synthesis, audio transcription, vectors, and decision-making. The model directory is fetched in real-time via /v1/models.

The image generation priority is as follows:

AIPDD → NewAPI → Local ComfyUI → Built-in → ImageGen

In other words, if AIPDD is configured, it is used first; if not, it falls back sequentially to NewAPI, local ComfyUI, built-in capabilities, and finally ImageGen.

API Key Configuration

This Skill is marked with requires_api_key=true. To use the AIPDD gateway, you must first prepare an AIPDD API Key.

  1. Open app.aipdd.work and complete registration.
  2. Enter “Global Settings → API Key Management” to generate an API Key.
  3. Provide the API Key in the conversation; the Skill will automatically save it to the skill package’s config.json for reuse next time.

The saving behavior here comes from the Skill’s own configuration flow; the specific field names should refer to the config.json within the skill package.

Installation and Enablement

The official installation entry is the SkillHub page:

https://api.skillhub.cn/org-bzwfrdvh/aipdd-char-view

SkillHub page address:

https://www.skillhub.cn/skills/org-bzwfrdvh/aipdd-char-view

The currently scraped data only provides the official addresses above and does not provide dedicated installation commands for Cursor, Codex CLI, or Claude Code. Since the Skill is based on the general SKILL.md format, it can usually be enabled from a directory or link according to the corresponding client’s Skill loading method. However, specific commands have not been verified in the materials and are not written here.

Typical Usage

The scraped data provides capability descriptions and API Key flows, but does not provide directly reproducible call examples, parameter lists, or output samples. Therefore, this article does not provide example commands to avoid writing unverified usage.

After accessing the official page, you can confirm the usage in the following order: first check if the version is 1.0.3, then confirm whether the AIPDD gateway is used, and finally pass the character settings according to the Skill invocation method supported by your actual client.

Suitable Scenarios and Notes

Suitable for the following situations:

  1. Short drama, comic drama, and animation projects need to maintain the same character’s face shape and proportions over the long term.
  2. Preparing input images for reference image workflows such as IP-Adapter.
  3. Creating training materials for character LoRA.
  4. Providing front/side/back three-view references for modeling or character art.

Points to note:

  1. Light gray and expressionless are deliberate restrictions. If the project requires character images with complete expressions, clothing, and lighting/shadows, this Skill is not a tool for final finished art.
  2. Facial bone structure, proportions, and age are forcibly locked, meaning it cannot be used to generate young versions, old versions, or exaggerated deformed versions of the same character.
  3. Image generation results depend on the model gateway; if AIPDD is unavailable, it will degrade according to priority, and the image style may be inconsistent under different gateways.
  4. There are discrepancies in the data regarding ratings and download counts: The Skill clues write a rating of 99,829 and 1,150 downloads; the SkillHub page scrape writes a rating of 100,000, 1,176 downloads, and 0 installs. When actually selecting, please refer to the real-time data on the SkillHub page, and do not treat the rating as a guarantee of model capability.

Conclusion

The value of this Skill lies in turning the issue of “inaccurate character faces” into a reproducible reference image generation process: a single 1:1 canvas provides facial features, multi-angle face shapes, and three-view drawings simultaneously, while forcibly locking the facial structure. Official address: https://api.skillhub.cn/org-bzwfrdvh/aipdd-char-view.