File size: 3,673 Bytes
20df93e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 | ---
tags:
- comfyui
- model-bundle
- image-generation
- video-generation
---
# Character Image Model Bundle
Deployment assets for the Character Image project, audited against commit `8f1d132` on 2026-09-07. This public repository contains the 25 referenced model files found on the source server, totaling 117,028,825,625 bytes (108.99 GiB).
## Directory Layout
| Directory | Contents | Files |
| --- | --- | ---: |
| `models/checkpoints/` | WAI v170, Qwen Rapid AIO v23, Pornworks Illustrious | 3 |
| `models/diffusion_models/` | Wan Remix V30 High/Low, two ZImage variants | 4 |
| `models/text_encoders/` | Wan UMT5 and Qwen 3 4B | 2 |
| `models/clip_vision/` | CLIP Vision H | 1 |
| `models/ipadapter/` | AnimeIPA and SDXL Plus Face | 2 |
| `models/loras/` | Six enabled LoRA dependencies | 6 |
| `models/vae/` | AE and Wan 2.1 VAE | 2 |
| `models/upscale_models/` | 4x AnimeSharp | 1 |
| `models/onnx/` | RIFE 4.9 ONNX | 1 |
| `auxiliary/censor/` | EraX, NudeNet, anime censor detector | 3 |
`model-manifest.json` lists each path, SHA256, byte size, and the template keys that reference it. Shared weights appear once. The model bytes and filenames are preserved from the source deployment.
## Download
Install `huggingface_hub`. This public repository can be downloaded without logging in. The following command downloads into the existing ComfyUI directory:
```bash
python3 - <<'PY'
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="ChatoyantAI/character-image-models",
local_dir="/root/ComfyUI",
max_workers=4,
token=False,
)
PY
```
The existing `models/` layout is directly compatible with ComfyUI. Re-running the download supports resuming an interrupted transfer.
## Restore Censor Dependencies
Two detectors are loaded through upstream HF cache paths, while NudeNet can load a local ONNX. After downloading:
```bash
python3 /root/ComfyUI/restore_censor_models.py --repo-dir /root/ComfyUI
export CENSOR_MODELS_DIR=/root/ComfyUI/auxiliary/censor
```
Keep the `CENSOR_MODELS_DIR` export in the service startup command, before ComfyUI starts. Run restoration as the service account, or supply its HF cache directory with `--cache-dir`. The script verifies and restores the EraX and anime detector cache entries at the recorded upstream revisions; their `refs/main` entries are set to those revisions. `CENSOR_MODELS_DIR` covers EraX/NudeNet, while the anime detector continues to use its HF cache.
The original node packages and Python dependencies are still required, including `ComfyUI-GenitalCensor`, `nudenet`, `ultralytics`, and `imgutils`.
## RIFE
`models/onnx/rife49_ensemble_True_scale_1_sim.onnx` is the input for `ComfyUI-Rife-Tensorrt`. Its node builds a TensorRT engine on the target machine when needed. This bundle contains no precompiled TensorRT engine.
## Coverage
Model dependencies were found for WAI_AnimeIPA, WAI_Qwen, WAI + Wan22 (image/text variants), Illustrious, WanImage, ZImageTurbo, and image/video censoring. Dependencies shared with other configured templates are also included.
Eighteen model references were absent from the source server and are not included: the legacy Pony checkpoint and five associated LoRAs; Wan Remix V20 High/Low; the Qwen next-scene LoRA; Monica/Trixie LoRAs and their ZImage NSFW LoRA; and six legacy video action LoRAs. Those templates require additional assets. Legacy Pony templates also require missing Comfyroll nodes and a compatible CLIP discovery filename.
Only model files and deployment metadata are included. Application configuration, credentials, user media, and logs are excluded. See the manifest for the exact uploaded set.
|