[docs] Prompting

[stevhliu]

Reworks the "Prompt techniques" guide to include better and clearer prompt writing instructions and cleans up the prompt weighting section a bit.

Also cleans up the "Batched inference" docs and includes example outputs.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[stevhliu]

@asomoza, is this still a useful or popular technique? Seems like quite a bit of work to code up the prompt enhancer (coming up with the words and styles) when you can just write a prompt with those same words in the first place.

No strong opinions either way, just looking for opportunities to streamline the docs a bit more :)

[asomoza]

this one in particular is not popular now, this was initially a technique used in Fooocus which made it really impressive back in the days, but today you can use very small llms locally or online to do the same and better, we have come a long way, we should probably remove this.

Instead I suggest to link some good models or we can even link to a search like this one: https://huggingface.co/models?sort=downloads&search=prompt+enhancer

There are also some models that have their own prompt enhancer like HunyuanImage-2.1, so this technique is still relevant but now it's being done by the researchers and model owners right out of the box instead of the users.

[sayakpaul]

@asomoza could you give this a review?

[asomoza]

thanks, looks great, I answered the question and left an additional comment

[asomoza]

this one in particular is not popular now, this was initially a technique used in Fooocus which made it really impressive back in the days, but today you can use very small llms locally or online to do the same and better, we have come a long way, we should probably remove this.

Instead I suggest to link some good models or we can even link to a search like this one: https://huggingface.co/models?sort=downloads&search=prompt+enhancer

There are also some models that have their own prompt enhancer like HunyuanImage-2.1, so this technique is still relevant but now it's being done by the researchers and model owners right out of the box instead of the users.

[stevhliu]

Thanks @asomoza! Feedback has been addressed, let me know if there is anything else, otherwise I think we can merge 🙂

[asomoza]

LGTM, thanks!