Yolo26 version is newest.
The zip file contains two models of different sizes (S and L) and a simple workflow.
Basically, S is sufficient, but you can use whichever you prefer.
Mosaic Detection series is integrated. penis, pussy, and penetration can be detected simultaneously.
This model is Adetailer for i2i or inpaint. GameCG,ArtistCG,and other.
Detects a mosaic censorsip, and redraw. Use it in conjunction with Lora or prompt as you like.
Recommended for use with comfyui.
denoising strength: around 0.5 ⇒Adjust it so that you can draw it well.
It's important to write detailed prompts.
Differences from previous version ↓
Improved detection accuracy.
If the angle is from the front where the stomach can be seen, the detection range will be wider. !BBOX version only!
⇒Possible to drawing pubic_hair's on-off,stomach_bulge,x-ray,uterus, etc...
In that case,increase denoising strength.
!!Atention!!
This model was trained with yolo11, so don't forget to update ultralytics.
\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
モザイク検出シリーズは統合されました。ペニス、マンコ、挿入を同時に検出できます。
このモデルはi2iまたはinpaint用のAdetailerです。ゲームCG、アーティストCGその他に。
モザイク修正を検出して再描画します。好みのLoraやプロンプトと合わせて使ってください。
Comfyuiでの使用を推奨
denoising strength:0.5前後 ⇒うまく描画できるように調整してください。
プロンプトは詳細に書いてください。
以前のバージョンとの違い ↓
1.検出精度の向上
2.お腹が見える正面からのアングルの場合、検出範囲が広くなります。!BBOX versionのみ!
⇒陰毛の有無、腹ボコ、透視、子宮などの描画が可能に。
その場合、denoising strengthを上げてください。
!!注意!!
このモデルはyolo11でトレーニングしたため、ultralyticsのアップデートを忘れずに。
Description
FAQ
Comments (12)
Nice model! Could you update Bar Censor Detection?
I will do it. Please wait a moment.
Thanks, Do you think you might switch from YOLO to SAM3?
@wildcats From my testing with SAM3 (which hasn't been fine-tuned yet), it does a good job segmenting censor bars and mosaics. However, it struggles if the mosaic grid is too fine or if the censored area is very small.
I recommend setting the threshold to around 0.45–0.60. You might be able to lower it if you separate the detection for censor bars and mosaics.
Here are the prompts I used if you want to try them: black bar, white bar, black rectangle, white rectangle, black strip, white strip, censorship bar, redaction bar, privacy bar, mosaic, mosaic censorship, mosaic pattern, pixelated, pixelated area, pixel grid, small squares pattern, censorship grid, pixel blocks, small pixelated patch, digital mosaic
Also, I recommend using the ( yolain/ComfyUI-Easy-Sam3: A ComfyUI custom node package for [SAM3 (Segment Anything Model 3)](https://github.com/facebookresearch/sam3), providing powerful image and video segmentation capabilities with text prompts. ) node for this.
Run prompt: Prompt outputs failed validation:
Lora Loader Stack (rgthree):
- Exception when validating inner node: tuple index out of range
How to resolve errors?
Try updating that node using the manager, and restart.
Can I ask where the download link for the prompt node in the workflow is? I can’t find it in the manager Thanks.
Chibi-nodes? here → https://github.com/chibiace/ComfyUI-Chibi-Nodes
Thank you for sharing your workflow! I have a question and would really appreciate your guidance.
I imported your 4th image into my ComfyUI to get the workflow. In the VAE loader node, I see the model is set to noobaiXLNAIXL_epsilonPred11Version. However, after I downloaded all the required models (using the exact names as they appear in the workflow, and the latest mosaic detection v4.0) and tried to run it, I encountered an error saying that noobaiXLNAIXL_epsilonPred11Version has no valid VAE detected (VAE is invalid: None).
I'm not sure if I made a mistake somewhere. Could you please help me figure out what's wrong? Looking forward to your reply. 🙂
In your case, the checkpoint model is probably located in the VAE folder. Please check the data size of your model. Checkpoints and VAEs have the same name, so they're easy to confuse.
"noob" is just a model I often use; I'm not necessarily recommending it. Please use the model and VAE you like best.
