Character Consistency Image · skill Open Skill công khai

Character Consistency Image

A plug n companion for ACO Video Director. A simple way to create consistent character likeness to avoid the usual drift that occurs when creating avatars or characters so looks of you creations stay locked in.

@napefx

ShortDrama

31 3 15

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/1716469/2026-10-08/sj7ckb2v.png

# Character Consistency Image

## When to use
- The user provides one or more character reference images (photo, illustration, portrait, still) and wants a single high-resolution, character-consistent reference image they can hand to ACO Video Director or any video generator.
- The user wants multiple references blended into one canonical character look, upscaled to the highest resolution, made background-free so it can be placed in any setting, and rendered as a multi-angle character sheet.
- The user wants the character-consistency info (canonical look + instructions for downstream generators) carried alongside the image so identity stays locked.

## Core principle
The goal is a **canonical character reference image**, a **multi-angle visual character sheet**, and a **companion character-consistency document** that travel together. The platform cannot write binary metadata into an image file, so the consistency info is delivered as a sidecar document (the equivalent of seed-locking / `--cref` / face-consistency / motion-brush notes) that ACO Video Director or any video generator can read to keep the character identical. The canonical reference is made **background-free** (character isolated on a transparent/plain background) so it is reusable in any setting. Recipe: **blend references → upscale to highest resolution → remove the background → build a multi-angle character sheet → write the consistency document → confirm everything is in the library → hand everything to the user.**

## Steps
1. **Gather references.** Collect all character reference images the user provides. If the user gives no image, ask for one (a character reference is required).
2. **Blend multiple references (if >1).** When there are multiple references, merge them into one canonical character using the best image-to-image model:
   - Call `generate_image` with `model='i2i.hd'` (HD multi-reference, best quality), `input_urls` = all reference images, `orientation` matching the source (default `portrait` for vertical), `resolution='4K'`.
   - Prompt: describe the unified character in detail — face, hair, body, outfit, and any features that must stay consistent — and state "merge these references into one consistent character, keep the face and identity identical across all of them."
   - `name` = `<character>_reference`, `entity_type` = `characters`.
   - If the user only provided a single reference, skip blending and use that image directly as the canonical reference.
3. **Upscale to the highest resolution.** Run the canonical reference through `generate_image` with `model='i2i.hd'`, `input_urls` = [canonical reference], `resolution='4K'` (fall back to `2K` if the model clamps), prompt = "upscale this character reference to the highest quality, keep the face and identity exactly the same, do not change the character." `name` = `<character>_reference_4k`, `entity_type` = `characters`.
4. **Remove the background (make it reusable in any setting).** Run the upscaled canonical reference through `generate_image` with `model='i2i.hd'`, `input_urls` = [upscaled reference], `resolution='4K'`, prompt = "isolate the character and remove the background entirely — output the character as a clean cutout on a plain transparent/white background, no setting, no scenery, no ground, so the character can be placed in any environment; keep the face and identity exactly the same." `name` = `<character>_cutout`, `entity_type` = `characters`. This background-free cutout is the canonical reference to hand to any video generator.
5. **Build a multi-angle visual character sheet.** Generate a turnaround sheet showing the same character from multiple angles, using the canonical cutout as reference:
   - Call `generate_image` with `model='i2i.hd'` (best for grid/panel layouts), `input_urls` = [canonical cutout], `orientation` matching the source (default `portrait`), `resolution='4K'`.
   - Prompt: "create a character sheet / turnaround of this exact same character shown from multiple