Slide show function on search results. (or maybe any time)

[osirigunso]

IIB's image collection is so amazing.
I like to find images from my collection that follow a particular theme at any given time and look at that slideshow forever.
Now I we need is a little automation and it will be a great time.
Thank you for this wonderful software.

[zanllp]

Hey osirigunso, really sorry I missed your message in discussions! 😅 Your idea of thematic slideshows on loop is great - actually aligns perfectly with our new TikTok-style viewer (try it with arrows keys!). We'll expand this with auto-looping in the next update - might be exactly what you're looking for.

[osirigunso]

Hi zanllp, thanks for finding the post!
And, it's a dream come true to have exactly the functionality I desired!
It's so useful and maximizes the efficiency of the use of generated products.
It's going to be an indispensable part of my generative AI life.
I can't wait it.

[zanllp]

just click that button to enable this feature.

[osirigunso]

Hi zanllp, thank you so much! Now I can look at the beautiful illustrations until I'm satisfied. I am really satisfied.
As far as I can tell, IIB doesn't have the functionality to sort the search results, so if I could randomize the order in which they are displayed, it would always keep things fresh.
If this is an idea for the future I would be happy.

I know this is out of place, but I'm new to github and not a programmer, so here's a note of the improvements I've made to tool.py about comfyui for better prompt reading.

def get_comfyui_exif_data(img: Image):
    if img.format == "PNG":
        prompt = img.info.get('prompt') or img.info.get('populated_text')   #comfyui-ImpactWC

    elif img.format == "WEBP":
        exif = img.info.get("exif")
        split = [x.decode("utf-8", errors="ignore") for x in exif.split(b"\x00")]
        prompt_str = find(split, lambda x: x.lower().startswith("prompt:"))
        populated_str = find(split, lambda x: x.lower().startswith("populated_text:"))   #comfyui-ImpactWC
        if prompt_str:
            prompt = prompt_str.split(":", 1)[1]
        elif populated_str:
            prompt = populated_str.split(":", 1)[1].strip()
    if not prompt:
        return {}  

    def get_text_from_clip(idx: str) :
        inputs = data[idx]["inputs"]
        for key in ["populated_text", "text", "t5xxl"]:   #comfyui-ImpactWC
            if key in inputs and inputs[key]:
                text = inputs[key]
                if isinstance(text, list):
                    text = data[text[0]]["inputs"].get("populated_text", data[text[0]]["inputs"].get("text", ""))
                return text.strip()
        return ""

[zanllp]

For the ComfyUI parser improvements: thanks for sharing those details! Since you've already reverse-engineered their metadata format, would you mind opening a PR with:

Your improved tool.py snippet
A few sample images (you can attach to the PR)
That way we can properly test and credit your contribution.