FAQ & Troubleshooting

My output imgs look completely different to each other, like it's doing pure txt2img and not img2img

• Go to your webui settings --> look for this setting, and make sure it's not checked:
  

My videos look like a mess when I upload them to social media platforms. What to do?

• The social media platforms are compressing the uploaded videos, so if you upload one (at low res), it will be compressed even further resulting in the unrecognizable goo. To prevent it, please upscale the videos at least 'x2' in the Output tab before posting anywhere except for Discord.

ControlNet is not working!

While we try our best to keep up with the latest CN updates, Deforum is not responsible for the ControlNet extension. Please, report the bugs here instead https://github.com/Mikubill/sd-webui-controlnet/issues !

3D animation mode is not working. Only 2D works

• If you got the AttributeError: 'NoneType' object has no attribute 'get', it means that one of the 3D model files is either missing, or downloaded only partially. The fix is to manually download the models again, and putting both of them in the /models/Deforum folder.
  
• Download AdaBins_nyu.pt from here
  
• Download dpt_large-midas-2f21e586.pt from here

I'm getting a JSONDecodeError: Expecting ',' delimiter error

• You have a missing comma (,) after one of your prompts. Make sure all of the prompts EXCEPT for the last one have a trailing comma after them.
• Should look like this:
  {"0": "Cool prompt at frame 0","60": "Cool prompt at frame 60","80": "Cool prompt at frame 80","100": "Cool prompt at frame 100"}
  and NOT like this (note the missing comma after the second prompt):
  {"0": "Cool prompt at frame 0","60": "Cool prompt at frame 60" "80": "Cool prompt at frame 80","100": "Cool prompt at frame 100"}
  

I reported an issue and it has been discarded by a bot 😔

Most likely, it happened because it didn't contain the full crash report. Good bugreports bot only have the error message and the traceback, but also all the info about your environment and all the other extensions, extra networks, etc. you're using. If the devs won't know all the information, it may lead them to chasing a ghost and simply frustrate everybody. Remember, «the software is provided "as is", without warranty of any kind» and the devs are all volunteers. So if you want to have your problem fixed, please at least provide it an unambiguous description.

TL;DR: your log should start like this

PS C:\Users\kabachuha\stable-diffusion-webui> .\webui-user.bat
venv "C:\Users\kabachuha\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr  5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.3.1
Commit hash: b6af0a3809ea869fb180633f9affcae4b199ffcf
Installing requirements

Launching Web UI with arguments: --medvram --no-half-vae

...
...
...

My image looks too flat, coloring book-like or deep fried

Like this:

• Most likely, is was caused by setting the 'Seed behaviour' to 'Fixed'. If the seed is the same, denoising doesn't happen and the image gets covered with noise over time. Please, don't use this option unless you know what you're doing. For beginners, it's better to use any other seed behavior, the simplest is 'Iter'.

Is Deforum supported on MacOS and MPS devices?

No, sorry.

My animation becomes blurry over time and loses details

Like this:

Or:

• There are two main ways of fixing it: either lowering the strength schedule or heightening the noise multiplier schedule... or using ControlNet.

The most luxurious and the most efficient is ControlNet in Loopback-mode. This way it feed the each frame as a ControlNet base for the next frame, preserving the details in a subtle way.

As for the other two ways

• These values are a good rule of thumb, but be cautious when mixing them together

• Strength schedule for arbitrary number of frames can look like this: 0:(0.25 * cos((72 / 60 * 3.141 * (t + 0) / 30))**13 + 0.7)

Example of strength drop (set up manually by mechakull on Discord):

0: (0.55), 20: (0.6), 30: (0.6), 41: (0.3), 60: (0.6), 70: (0.6), 81: (0.3), 100: (0.6), 110: (0.6), 121: (0.3), 140: (0.6), 150: (0.6), 161: (0.35), 180: (0.6), 200: (0.6), 201: (0.3), 220: (0.6), 230: (0.6), 241: (0.3), 260: (0.6), 270: (0.6), 281: (0.35), 300: (0.6), 310: (0.6), 321: (0.35), 340: (0.6), 350: (0.6), 361: (0.3), 380: (0.6), 401:(0.3), 420: (0.6), 430: (0.6), 441: (0.35), 460: (0.6, 470: (0.6), 481: (0.3), 500: (0.6), 510: (0.6), 521: (0.3), 540: (0.6), 550: (0.6), 561: (0.3), 580: (0.6), 600:(0.3)

• Noise multiplier: 0: (1.07-(abs(sin(2*3.14*t/5+0))*0.02))

Example of increased noise multiplier:

My animation gains a too sharpened look, quickly. Something like a grid or a net appears

Like this:

• Lower Anti-Blur's amount_schedule or noise multiplier schedule if you have it enabled.

Deforum crashes with KeyError after I press 'Generate'. It seems it is related to the keyframes you added recently.

  File "/home/kabachuha/kml/kek/stable-diffusion-webui/extensions/deforum/scripts/deforum_helpers/animation_key_frames.py", line 9, in __init__
    self.angle_series = get_inbetweens(parse_key_frames(anim_args.angle), anim_args.max_frames)
  File "/home/kabachuha/kml/kek/stable-diffusion-webui/extensions/deforum/scripts/deforum_helpers/animation_key_frames.py", line 79, in get_inbetweens
    key_frame_series[0] = key_frame_series[key_frame_series.first_valid_index()]
  File "/home/kabachuha/.local/lib/python3.8/site-packages/pandas/core/series.py", line 982, in __getitem__
    return self._get_value(key)
  File "/home/kabachuha/.local/lib/python3.8/site-packages/pandas/core/series.py", line 1092, in _get_value
    loc = self.index.get_loc(label)
  File "/home/kabachuha/.local/lib/python3.8/site-packages/pandas/core/indexes/range.py", line 396, in get_loc
    raise KeyError(key)
KeyError: None

• Go to webui's core folder and remove ui-config.json. This action will clear the cache and it often solves this issue (especially if no one else seems to have it, as each other's cache forms in a different way)

How to create a transition between any pre-existing images

There are two main ways to do so:

• Use the 'Guided images' mode in the extension.
• Interpolate the pics using a frame interpolation engine in the 'Output' tab.

Guided images

The 'Guided images' mode slowly blends in the target images into the usual Deforum animation process using a complicated math formula, before the diffusion starts to remake the images. This way the images 'guide' the AI to morph the frames being made towards the intended shapes.

Frames used:

Frame interpolation

There are two main interpolation engines: RIFE and FILM. FILM is slower and demands more resources, but it's worth it.

Insert the pics or a video into the output tab, select the needed settings and interpolate the pics of your choice, even not AI-generated!

Frames used:

Naturally, interpolate your video before upscaling or else it will take a ton of time.

Can I generate the videos directly on the Discord server?

As of now, there are NO official Deforum-generation Discord bots. That being said, sorry, but you'll have to generate the images on your own PC or in a cloud.

... Unless you donate us enough money to host one, wink-wink

Preview window is stuck/ not being updated during animation generation

This is a known issue with no known fix as of January 19th 2023. Please wait for a fix. Issue can be tracked Fixed with an update of the core auto. If you have the issue, upgrade your webui.

Where is the "Hires. fix"? Is there one? Did you remove it? Why?

• There was a button, but it turned out what it applied only to the first frame of the animation, thus it used to create only an illusion of usefulness. We decided to remove it not to confuse the users and to free the UI space.

How do I protect my animations from duplicating the content then, such as making multiple heads or even splitting the horizon?

Please, render it on lesser resolution. (the greater dim should be not much more than 512 for SD 1.X. 728 for 2.X) You can always upscale your videos using Deforum's internal upscalers in the Output tab.