AI Image Creation

[David L]

Looks like Geddy's image should have had porn on the PC 
Alex is definitely cultivating a bit of a 'whisky nose', like Neil's but not quite as well developed. May have been 'created' by AI but I think not 
That image also looks 'sans hairpiece' so definitely candid 

[Slim]

Alex is definitely cultivating a bit of a 'whisky nose', like Neil's but not quite as well developed. May have been 'created' by AI but I think not 

Not sure what you mean? I asked ChatGPT for an image of Alex Lifeson repairing a toaster in a garage, looking over his shoulder to the camera.

[David L]

Alex is definitely cultivating a bit of a 'whisky nose', like Neil's but not quite as well developed. May have been 'created' by AI but I think not 

Not sure what you mean? I asked ChatGPT for an image of Alex Lifeson repairing a toaster in a garage, looking over his shoulder to the camera.

I mean:
If you asked AI to create a 'current' image of Alex, it may take into account his propensity to indulge quite heavily in alcohol and give his nose a bit of a purplish bloom. Something akin to 'The Professor's' whisky-induced hooter.
Or it may have just used a stock image, because he's definitely got the start of a pint-swiller's proboscis. Possibly the fault of Henderson's brewery or hanging at Geddy's.

[Slim]

It does use stock images in a sense, because it will have been trained on multiple images of him. But it's unlikely that it's a simple incorporation of a single stock image. I think. Actually I'll ask it ..

The image you provided of Alex Lifeson earlier. It was quite realistic. Is the particular representation of Alex' head possibly derived from a stock image? Or is it a reimagination based on your training data, which (presumably) contains multiple images of him - so you constructed a new image based on information points from those?

Great question—and you're right on both counts in a way, but here's the breakdown:

The image I generated of Alex Lifeson isn't taken from a specific stock photo or an existing image. Instead, it's a synthetic reconstruction—a reimagining based on a large number of visual and textual data points from training. The model has been trained on many images of public figures like Lifeson, including how he looks at various ages, his general facial features, hair, build, and common expressions. So when you ask for a realistic image of "modern-day Alex Lifeson," the model draws from that overall pattern and creates something new that resembles him in a plausible, coherent way—kind of like a digital impressionist who's seen a bunch of photos but isn't copying any one in particular.

It's a bit like what a sketch artist might do if they'd studied someone's face from different angles and then drew them doing something the artist had never seen before.

The result might feel a bit uncanny at times—realistic, but not quite identical to a specific known photo. If you put the generated image next to real photos, you'd likely spot differences, even though the vibe and features might be spot-on.

[The Picnic Wasp]

Scary!

[David L]

It does use stock images in a sense, because it will have been trained on multiple images of him. But it's unlikely that it's a simple incorporation of a single stock image. I think. Actually I'll ask it ..

The image you provided of Alex Lifeson earlier. It was quite realistic. Is the particular representation of Alex' head possibly derived from a stock image? Or is it a reimagination based on your training data, which (presumably) contains multiple images of him - so you constructed a new image based on information points from those?

Great question—and you're right on both counts in a way, but here's the breakdown:

The image I generated of Alex Lifeson isn't taken from a specific stock photo or an existing image. Instead, it's a synthetic reconstruction—a reimagining based on a large number of visual and textual data points from training. The model has been trained on many images of public figures like Lifeson, including how he looks at various ages, his general facial features, hair, build, and common expressions. So when you ask for a realistic image of "modern-day Alex Lifeson," the model draws from that overall pattern and creates something new that resembles him in a plausible, coherent way—kind of like a digital impressionist who's seen a bunch of photos but isn't copying any one in particular.

It's a bit like what a sketch artist might do if they'd studied someone's face from different angles and then drew them doing something the artist had never seen before.

The result might feel a bit uncanny at times—realistic, but not quite identical to a specific known photo. If you put the generated image next to real photos, you'd likely spot differences, even though the vibe and features might be spot-on.

Interesting response. The largest proportion of images that it referred to must've been without the syrup because it seems to have 'aggregated' those without and those with and came up with an image that looks far more like he's sporting his own hair rather than the full rug. I think it's definitely a bug with AI that needs to be ironed out (the 'rug' bug?).
Try Ian Paice dancing with a Ladyboy in a laundrette or, better still, Richie Blackmore playing mousetrap with the Pope. We'd know for sure if the 'rug' bug needs fixing.

[Slim]

Highslide Image Viewer

[The Picnic Wasp]

Highslide Image Viewer

AI obviously forgot about the cool Chopper gear change lever. Mine had two as it was a 5-speed, which was quite rare. Now that my memory has been jogged I've remembered that Raleigh also produced a drop handlebars racing version. I don't remember ever seeing one other than in Dales Cycles in their original Glasgow store. Great bike shop Dales. Bought my e-bike there last year. Really knowledgeable, friendly staff and still family owned and run. I was quite surprised at their stock levels on my last visit. Must be a huge financial commitment.

[Slim]

Yes it's odd that text generation is so unreliable. You'd think that it would be the simplest aspect of the task. But it's a preview version so hopefully it will get ironed out.

I tried a few text-based images with Gemini last night and noticed that this was much improved:

Highslide Image Viewer

Highslide Image Viewer

I pointed this out to the AI, which replied:

Highslide Image Viewer

[Slim]

.. however I've just tried the bing.com image generator, and it still messes up text.

[Fishy]

Highslide Image Viewer

Thought I'd give co pilot a go..

[Thenop]

Highslide Image Viewer

Thought I'd give co pilot a go..

Alex before his stomach stitch up then

[Slim]

Highslide Image Viewer


https://huggingface.co/spaces/multimodalart/Qwen-Image-Edit-Fast

"Make the guitar gloss black" 

Highslide Image Viewer

[Thenop]

So now you know you should stick with the burst.

[David L]

"Remove the scratch plate"