Why can I only have 75 characters after updating the latest version?

[kidps2]

Only 75 characters can't be used normally. Some keywords have more than 20 characters, some artists have more than 20 characters, and some keywords have a series of 50 characters. The limit of 75 characters can't be used normally

[kidps2]

Did I use up any ghostwriting? Or is the old version also limited to 75, but the new version is displayed?

[dovezp]

It has to do with the tokenize limit for input characters. Long story short, Stable Diffusion has a limit for input text length.

In general, a prompt must be 75 tokens or less, anything above will result in "prompt truncated after tokenization". Your prompt is case insensitive. There's about 30,000 tokens understood by the AI, this means it won't know about some weird word unused since the 1600s. A token is roughly a word, a punctuation, or a Unicode character.

You can preview some common prompts here: https://promptomania.com/stable-diffusion-prompt-builder/

This is where Textual Inversion and saving existing tokenizer parameters as Styles come into play.

Styles Example:

1. Enter a positive prompt: cute kitten
2. Generate
   

cute kitten
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 3270996573, Size: 512x512, Model hash: 9453d3e5

3. Create Style, name it cute_kitten
4. Enter new positive prompt: Giovanni Fattori
5. Apply new cute_kitten Style
   
6. Generate
   

Giovanni Fattori, cute kitten
Steps: 20, Sampler: Euler a, CFG scale: 27.5, Seed: 1676737396, Size: 512x512, Model hash: 9453d3e5

Textual Inversion Example:

1. Find a concept-image model / embed model. Personally I look at https://huggingface.co/sd-concepts-library

2. Download your model of choice, for example https://huggingface.co/sd-concepts-library/852style-girl

3. Rename {learned_embeds}.bin to the class name that is inside token_identifier.txt. For example, I copy <852style-girl> and name the bin file, 852style-girl.bin

4. Move the newly named bin file to the \embeddings\ directory

5. Now boot up webui and lets call this in the prompt. For our example: <852style-girl>, photorealistic, oil painting, vintage

6. Click generate. You should notice a prompt in your log window Loaded a total of {{x}} textual inversion embeddings.. Give it some time to process.

7. Profit
   

<852style-girl>, photorealistic, oil painting, vintage
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 3895043234, Size: 512x768, Model hash: 9453d3e5

[dovezp]

You can manually tweak the ui.py if you would like to mess with the tokenizer functions, and navigate backwards from there to see what is happening but I personally don't recommend it.

https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/modules/ui.py#L348