-

2026-01-02 17:45:18

iolaire@masto.ai

I'm looking for some feedback on AI-generated #AltText. Please, are you willing to provide feedback on the quality and if it’s helpful for you personally? I have a PixelFed profile that I'd share for examples. It’s all #photographs. I've been working on tools to do it for a while.

2026-01-03 00:06:29

Jupiter Rowland

jupiter_rowland@hub.netzgemeinde.eu

@iolaire I've pitted an image-describing LLM AI against my own 100% hand-written image descriptions twice so far. I have first described an image myself, twice even, with a "short" description for the alt-text and a long, fully detailed description with text transcripts and all necessary explanations for the post text.

However, I'm always at an unfair advantage. My images are renderings from very obscure 3-D virtual worlds. LLMs know next to nothing or actually nothing about these worlds whereas I dare say I'm an expert on them. An AI couldn't even tell whether the image is from a game or from a virtual world, much less which virtual world. I can not only exactly pinpoint where the image was taken (which place on which sim in which grid), but also explain the location and these virtual worlds in general.

Besides, an AI would describe the image by examining the image. I describe my images by going in-world and looking at the real deal instead of at the image of it. I can see everything at a vastly higher resolution. I can transcribe text that is so tiny in the image that it's invisible. I can even look around obstacles and see what's behind them if necessary. No LLM AI can do any of this.

(1/6)

#Long #LongPost #CWLong #CWLongPost #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta #AI #AIVsHuman #HumanVsAI

2026-01-03 00:06:53

Jupiter Rowland

jupiter_rowland@hub.netzgemeinde.eu

@iolaire Allow me to give you an example.

This is the image I'm talking about: https://hub.netzgemeinde.eu/photos/jupiter_rowland/image/b1e7bf9c-07d8-45b6-90bb-f43e27199295 (linked instead of embedded so I don't have to go through the hassle of having to describe it right here right now).

This is the thread in which I've posted the image before, including image descriptions, also including a comment with the AI description and an analysis of the AI description in comparison with my own descriptions: https://hub.netzgemeinde.eu/item/f8ac991d-b64b-4290-be69-28feb51ba2a7 (yes, this is part of the Fediverse; it's on the same Hubzilla channel that I'm commenting from right now).

(2/6)

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta #AI #AIVsHuman #HumanVsAI

2026-01-03 01:10:34

iolaire

iolaire@masto.ai

@jupiter_rowland thank you. I’ll look at your examples from my desktop tomorrow.

I was using LLaVA locally via Ollama. Since I moved to AWS Bedrock Nova Lite since it’s hosted and extremely affordable at about 2/10 of a cent per image. It’s fairly consise but honestly I’ve not tried to make it return longer descriptions.

Example https://masto.ai/@iolaire/115828239039871251

iolaire schrieb den folgenden Beitrag Sat, 03 Jan 2026 01:04:26 +0100

A close-up view of a rock surface covered with various types of lichen, including yellow, orange, and white patches, set against a rocky background. (AI-generated Alt text)

A photo with alt text. Using AWS Bedrock Nova Lite. Jerez de la Frontera, Andalucia, Spain 2025

#Lichen #NaturePhotography #RockSurface #Outdoors #Wildlife