bug : AttributeError: 'FreeTypeFont' object has no attribute 'read'

[onewolf24]

I try to run train an embedding. when it get to the point where it should save the embedding it said the finished training. the training should not finished yet and I open look at the CMD and it said.

stable-diffusion-webui\venv\lib\site-packages\PIL\ImageFont.py", line 232, in load_from_bytes
self.font_bytes = f.read()
AttributeError: 'FreeTypeFont' object has no attribute 'read'

I have tried deleting the venv file. this started happing on the 26th.

[TheMoonyMarigold]

I'm a newbie to SD and pretty unfamiliar with coding/tech in general, but I am having the same issue. I was successfully running training exercises over the last couple of weeks with stable diffusion (through google colab) until I encountered this bug (beggining today). The training always "finishes" one step before the n variable I've set where n = the step interval at which I've asked it to save a copy of both the image and the embedding. Pretty lost as to how to fix, let me know if you find a solution!

I set the max steps to 34000 (I have 34 training images), and I set n=136 for both the image and embedding logs. It says it's preparing the datasets and runs through 4 epochs then gives me this message in the UI: "Training finished at 135 steps.
Embedding saved to my file path name" and here's the code it produces:

Preparing dataset...
100% 34/34 [00:38<00:00, 1.12s/it]
Training textual inversion [Epoch 3: 34/34] loss: 0.0570441: 0% 135/34000 [01:30<6:51:46, 1.37it/s]
0% 0/20 [00:00<?, ?it/s]
5% 1/20 [00:01<00:24, 1.31s/it]
10% 2/20 [00:01<00:12, 1.42it/s]
15% 3/20 [00:01<00:08, 1.98it/s]
20% 4/20 [00:02<00:06, 2.42it/s]
25% 5/20 [00:02<00:05, 2.74it/s]
30% 6/20 [00:02<00:04, 3.00it/s]
35% 7/20 [00:02<00:04, 3.19it/s]
40% 8/20 [00:03<00:03, 3.33it/s]
45% 9/20 [00:03<00:03, 3.43it/s]
50% 10/20 [00:03<00:02, 3.50it/s]
55% 11/20 [00:04<00:02, 3.56it/s]
60% 12/20 [00:04<00:02, 3.56it/s]
65% 13/20 [00:04<00:01, 3.60it/s]
70% 14/20 [00:04<00:01, 3.62it/s]
75% 15/20 [00:05<00:01, 3.64it/s]
80% 16/20 [00:05<00:01, 3.64it/s]
85% 17/20 [00:05<00:00, 3.59it/s]
90% 18/20 [00:05<00:00, 3.60it/s]
95% 19/20 [00:06<00:00, 3.62it/s]
100% 20/20 [00:06<00:00, 3.07it/s]
Traceback (most recent call last):
File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/textual_inversion/textual_inversion.py", line 612, in train_embedding
captioned_image = caption_image_overlay(image, title, footer_left, footer_mid, footer_right)
File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/textual_inversion/image_embedding.py", line 150, in caption_image_overlay
font = ImageFont.truetype(textfont, fontsize)
File "/usr/local/lib/python3.10/dist-packages/PIL/ImageFont.py", line 855, in truetype
return freetype(font)
File "/usr/local/lib/python3.10/dist-packages/PIL/ImageFont.py", line 852, in freetype
return FreeTypeFont(font, size, index, encoding, layout_engine)
File "/usr/local/lib/python3.10/dist-packages/PIL/ImageFont.py", line 215, in init
load_from_bytes(font)
File "/usr/local/lib/python3.10/dist-packages/PIL/ImageFont.py", line 195, in load_from_bytes
self.font_bytes = f.read()
AttributeError: 'FreeTypeFont' object has no attribute 'read'

Applying optimization: sdp... done.

[chromesun]

I get the same error during TI training at the same place, N-1 when image save=N. System is Win10, automatic1111 build 20ae71f (from 2023 May 27).
Reverted to my last-known-good build (i.e. last build I managed a successful TI training): 89f9faa (2023 May 14).
TI training works again.
Given what onewolf24 and TheMoonyMarigold say above perhaps a build somewhere in between 14th and 25th would also work. Don't have time at the moment to narrow it down.

[TheMoonyMarigold]

@chromesun

Thanks so much for the input! My apologies in advance for the simpleton question, but how do I revert to a previous build running the fast_stable_diffusion_AUTOMATIC1111.ipynb notebook on Google Colab (from https://github.com/TheLastBen/fast-stable-diffusion)?

I'm not even sure how to tell what build I am currently running--again, I am new to this whole world, but learning!

Does it have something to do with changing the number/letter sequence that I've highlighted in the photo below, perhaps (d2VidWk)?

[chromesun]

Sorry TheMoonyMarigold but I've never run a Colab notebook, so I have no idea how they work. I wonder if the Use_Latest_Working_Commit checkbox would help? AFAIK the 89f9faa build that's working locally for me is v1.2.1, whereas the 20ae71f build that isn't working for TI is v1.3.0 of Automatic1111's webui. Perhaps it's worth trying that checkbox?
If not, I hope a Colab expert can jump in here.

[TheMoonyMarigold]

@chromesun Thanks for the reply! I will post if I end up figuring out how to revert to a previous build in Colab. I tried checking the box as you suggested, but the notebook wouldn't actually run properly at all with it checked.

[onewolf24]

how do I use an older version of sd , until this is fixed ?

[TheMoonyMarigold]

@onewolf24 I am testing out replacing the number I highlighted in my repsonse above with a number from this page: https://github.com/AUTOMATIC1111/stable-diffusion-webui/commits/master/CHANGELOG.md (my last successful running of SD was on the 25 of May so I'm going with May 21st).

I'll let you know if it works.

[TheMoonyMarigold]

okay so replacing it in that location produces an error, but I do think we use that number somehow. stay tuned.

[chromesun]

@onewolf24 I use ZIPs downloaded from
https://github.com/AUTOMATIC1111/stable-diffusion-webui
(the green Code button on that page)
so I just overwrite my a1111 folder with a previous zip. If you're using git then I'm sure I've seen mention of how to revert in either the Discussions or Issues forums here. Search for "revert" maybe? Sorry I can't be more useful.

[chromesun]

@onewolf24 I should have said, if you do use ZIPs you can get an older one by scrolling through the commits page(s) and then using the <> button to the right of the commit hash to get to the repository at that state. Then use Code button and "Download ZIP".
I'd paste a screen snippet but this site won't let me.

[chromesun]

@onewolf24 This seems relevant for the git method to revert:
#9039

[onewolf24]

I could try it and still had the same error

[onewolf24]

I found out the unchecking "Save images with embedding in PNG chunks" and the trainning work again. A bug report was submitted #10778 and they are working on fix

[TheMoonyMarigold]

@onewolf24 Thank you!

[chromesun]

@onewolf24 Interesting and useful! Once my current training run is finished I'll try that out.