Questions about the input
该模型似乎对输入格式有较为严格的要求，当我输入了一个过长的文本时，生成的图像其画面直接崩坏了，有时只能用初具人形来形容，效果非常差劲，想办法将文本缩减到了原来的一半反而变得好得多，这方面有什么文档之类的总结过吗？
This model seems to have fairly strict requirements for the input format. When I inputted a very long text, the generated image's visuals completely broke down—sometimes it could only be described as barely humanoid, with extremely poor results. Reducing the text to half its original length by various means made it much better. Is there any documentation or summary on this aspect?
Examples please, how long are we talkin'? I've had no issues yet and my prompts are longer than the average.
Examples please, how long are we talkin'? I've had no issues yet and my prompts are longer than the average.
about 109 words
Examples please, how long are we talkin'? I've had no issues yet and my prompts are longer than the average.
about 109 words
natural language
about 109 words
That's not an example, I have prompts well into 170+ words which get processed without issue, and I've tried synthetic prompts with 360+ words, the word count doesn't really matter. Make sure your prompts have no internal contradictions and are structured properly, then you can probably use 1000+ words.
Edit: I asked for an example because usually the prompt is at fault, token count hasn't been an issue since SD1.4.
about 109 words
That's not an example, I have prompts well into 170+ words which get processed without issue, and I've tried synthetic prompts with 360+ words, the word count doesn't really matter. Make sure your prompts have no internal contradictions and are structured properly, then you can probably use 1000+ words.
Edit: I asked for an example because usually the prompt is at fault, token count hasn't been an issue since SD1.4.
However, such a situation did actually occur, and I'm not sure what happened. The prompt I used was NSFW, so I probably can’t share it directly. But the process of halving this prompt to make it "usable" simply involved sending it to an LLM with the instruction: "Shorten this sentence without altering its core meaning."
Maybe the issue lies in the fact that I first had the LLM translate my native language into English, but during that process, the LLM failed to accurately understand my intent, resulting in a logical error in the English prompt after translation. Then, when I had the LLM shorten that prompt, it suddenly spotted the error and fixed it on the spot. If the truth is this ridiculous, I have nothing to say...
Post a damn example buddy...