FaceFusion 3.2.0 — How to Disable NSFW Filter
So, the old method using PROBABILITY_LIMIT doesn't work anymore in FaceFusion 3.2.0 (based on this update 28-04-2025).
But here's a new way that worked for me — no coding skills needed, just follow these steps:
🚧 Steps:
Open this file:
- Go into the
facefusionfolder - Then open the file
facefusion/content_analyser.pylike in this github using notepad or any text editor. - In pinokio:
pinokio/api/facefusion-pinokio.git/facefusion/facefusion/content_analyser.pyor just click the View Folder:
Find this part in the file:
-
Look for the line that starts with:
def detect_nsfw This section is where FaceFusion checks if content is NSFW (Not Safe for Work).
You'll see several lines below it that are indented (moved slightly to the right). That's normal.
Find this line:
keep_indices = numpy.where(nsfw_scores_raw > 0.2)[0]
Change it to:
keep_indices = numpy.where(nsfw_scores_raw < 0.0)[0]
- Mind the
<Symbol. so, the changes are from> 0.2to< 0.0.
✅ That's the only change.
⚠️ Important:
- Don't change the spacing in front of the line.
- Python uses indentation (spaces at the beginning of lines) to group code.
- If you break the spacing, the app might crash with a "IndentationError".
- Make sure you use
< 0.0, not just< 0, or it might not work right.
Save and Restart
- Save the file after editing.
- Restart FaceFusion if it's already running.
⚠️ Reminder
This disables the NSFW filter — so be responsible with what you use. (I'll be damned, really? I'm too much of a hypocrite for this..)
This is just from trial and error — no guarantee it'll always work.
Top comments (24)
keep_indices = numpy.where(nsfw_scores_raw < 0.0)[0] ==>>> No works for me, but....
keep_indices = numpy.where(nsfw_scores_raw > 0.0)[0]==>>> Works very well , Thanks from France ;)
muito obrigado, e muito sucesso!!!
Agradeço! E para você também!
solution:
def detect_nsfw(vision_frame : VisionFrame) -> List[Score]:
return [] # Disabled NSFW check
still have error
Analysing: 100%|================================================================ | 4524/4537 [00:15<00:00, 283.78frame/s, rate=0]
Traceback (most recent call last):
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git.env\lib\site-packages\gradio\queueing.py", line 625, in process_events
response = await route_utils.call_process_api(
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git.env\lib\site-packages\gradio\route_utils.py", line 322, in call_process_api
output = await app.get_blocks().process_api(
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git.env\lib\site-packages\gradio\blocks.py", line 2146, in process_api
result = await self.call_function(
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git.env\lib\site-packages\gradio\blocks.py", line 1664, in call_function
prediction = await anyio.to_thread.run_sync( # type: ignore
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git.env\lib\site-packages\anyio\to_thread.py", line 56, in run_sync
return await get_async_backend().run_sync_in_worker_thread(
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git.env\lib\site-packages\anyio_backends_asyncio.py", line 2470, in run_sync_in_worker_thread
return await future
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git.env\lib\site-packages\anyio_backends_asyncio.py", line 967, in run
result = context.run(func, *args)
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git.env\lib\site-packages\gradio\utils.py", line 884, in wrapper
response = f(*args, **kwargs)
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\uis\components\instant_runner.py", line 82, in run
create_and_run_job(step_args)
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\uis\components\instant_runner.py", line 97, in create_and_run_job
return job_manager.create_job(job_id) and job_manager.add_step(job_id, step_args) and job_manager.submit_job(job_id) and job_runner.run_job(job_id, process_step)
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\jobs\job_runner.py", line 11, in run_job
if run_steps(job_id, process_step) and finalize_steps(job_id):
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\jobs\job_runner.py", line 72, in run_steps
if not run_step(job_id, index, step, process_step):
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\jobs\job_runner.py", line 58, in run_step
if job_manager.set_step_status(job_id, step_index, 'started') and process_step(job_id, step_index, step_args):
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\core.py", line 323, in process_step
error_code = conditional_process()
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\core.py", line 340, in conditional_process
return process_video(start_time)
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\core.py", line 418, in process_video
if analyse_video(state_manager.get_item('target_path'), trim_frame_start, trim_frame_end):
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\content_analyser.py", line 102, in analyse_video
if analyse_frame(vision_frame):
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\content_analyser.py", line 77, in analyse_frame
nsfw_scores = detect_nsfw(vision_frame)
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\content_analyser.py", line 115, in detect_nsfw
temp_vision_frame = fit_frame(vision_frame, model_size)
File "D:\AISoftware\pinokio\api\facefusion-pinokio.git\facefusion\facefusion\vision.py", line 243, in fit_frame
height, width = vision_frame.shape[:2]
AttributeError: 'NoneType' object has no attribute 'shape'
The error occurs before the nsfw detection at
temp_vision_frame = fit_frame(vision_frame, model_size). it's not even invoking thekeep_indices = numpy.where(nsfw_scores_raw < 0.0)[0]yet.It's irrelevant to my topic.
you can contact the developer. because, AFAIK, in this 3.2.0, they use new method of nsfw detection that I believe raise this error.
full changelogs
you should report the bug to them.
Load up the content_analyser.py file as above, search for 'def analyse_frame' and edit the one found on line 70 with the following;
def analyse_frame(vision_frame : VisionFrame) -> bool:
return False
Save and reload FF
It doesn't seem to work, even using 3.2.0.
Not even trying to do NSFW.
Just messing around with family photos and videos right now, and even someone on a rollercoaster can't be swapped.
You might not be following his instructions properly. Perhaps you could send him a message and ask for personal guidance. Don't forget to buy him a coffee.
anyway, thanks @ijash !
Don't see how I'm not following the instructions. While I'm not the smartest person in the room, they seem simple enough for someone who was able to install FaceFusion in the first place.
Perhaps I'm wrong, but here are screenshots. (I purposely blurred the faces in the original photo, but you can see that the whole picture is blurred in the preview. This happens even if I put in a new subject for input)
You need to change the ">" to a "<" as well; it looks like you forgot to change it.
Fudge...
I feel SUPER stupid right now.
Thanks for that, I DID totally miss that and just looked at the numbers.
Ugh...
So... how is it now?
btw, thank you @kenny_9999995 & @nanonick . appreciate it.
we need a video tuto, iam to stupid for that, everytime when i change the two to a Zero , the complete text disappears
you use windows?
yes windows 11
Sorry, I use linux. even if i did make it, there'll be some visual difference. but the underlying concept is the same. I'll see what I can do, meanwhile, please wait. Because I'm not too keen in video editing.
Can you just comment out the whole function and return a safe score value? The NSFW analyzing function still runs, which slows the overall process down, why not just avoid it altogether?
i am aware of that. but i'm no python expert, in my best guess, every time this function called, it expect a certain type of value. that's why I don't really avoid it.
may be it can be substituted with empty array.. maybe..
not working
give me your code screenshot.
Is there a way to run it on colab without errors? Thx in advance 🙏
have you tried?