I started out with ChatGPT, but nowadays I mostly use Google's AI Studio. Both are free; Google's AI Studio tends to be less captious about possible improprieties.
NanoBanana (Google AI)
This link will no doubt be superseded soon.
DeeVid (paid) has also been great.
I apologize for the crudity of some of my efforts. Quality will improve, not least through advances in image LLMs. Also I am new to this, and any advice will be gratefully received.
Here are my notes, so far.
Getting satisfactory results requires (inter alia):
(a) eagerness to experiment,
(b) much time + patience, and
(c) some degree of ruthlessness, in rejecting subpar outputs.
Prompts should be refined, but know that a single prompt may yield vastly different results depending on which AI thread is assigned your request. Try, and try again.
Your role in the cleverness chain isn't just to issue requests: you also have options in dealing with the output. Retain images that are only partly successful, because you might be able to composite them into something useful. And sometimes our old "electronic surgery" tools still come in handy, like Magnetic Lasso and Liquify in Photoshop. A game-changer is that you can ask AI to render only the background, without the protagonist(s), and then use that as a separate layer in your photo editing app. This lessens your reliance on the tell-tale Clone Stamp Tool.
Lastly (for now):
(a) when in doubt about an image, look for reasons to discard it rather than reasons to keep it, and
(b) be open to happenstance: the sweetest delights are often accidental ones.
pixelskylines
2087
jbbartram-illu
yousharknotpass
gummygunk
areudirty
dededos
jacobvanloon
prguitarman
theartofmadeline
