Been working on AI pictures v2

[TheGreatMoney]

(This is an AI created image not me or my family)




What do you think the next ones should be?

[Undertaker]

I like your style. But we've come much further. Some other sites can do that as a full video/animation

[Ria99]

What else can you do?

[Blake_73]

Moonshot

[Kev94]

Turn it onto another planet

[Harbour]

(This is an AI created image not me or my family) [img]https://qday.f

Cheers for that. Nice one

[DarkEnergy]

Not sure I am fully with you on that one. Good stuff

[Phil]

So these are really brilliant work

Much better than my instructions 

[IronFist21]

Beautiful pictures. 

[Fan22]

These are really striking, especially the way the lighting gives them a slightly dreamlike quality without making everything look completely unreal. For the next batch, I would try changing the mood rather than just changing the subject. Keep the same general visual style, but do one set at dawn, one in harsh midday light, and one with strong artificial lighting.

That would make it interesting to see how much of the look comes from the subject and how much comes from the lighting choices.

[SingularityNodeKettle]

The next ones could tell a tiny story. Maybe three images showing the same character or place before, during, and after some event. AI images are often strongest visually when there is a clear concept behind them, and a sequence gives people something to look for beyond individual pretty pictures.

Even something simple like a lonely roadside diner, then the diner during a storm, then the same place years later could work really well.

[Baz_26]

One thing worth experimenting with is imperfect composition. A lot of generated images tend toward the polished poster look, where everything is conveniently dramatic and perfectly placed.

Try asking for something more observational: a subject partially blocked by a doorway, an awkward crop, someone looking away from the camera, or a scene where the main action is happening in the background. Those little imperfections can make an image feel much more photographic.

[Compiled Wolf]

The reflections and atmospheric depth are doing a lot of work here. For the next round, I would lean into weather: fog, rain on glass, wet pavement, snow, or dusty sunlight. Those conditions give you layers in the image and make the lighting much more interesting.

Rainy nighttime streets in particular could be fun because you can play with reflections, neon, headlights, and silhouettes without needing the image to explain everything.

[Paige_26]

Could you try a series where the same scene is generated in completely different artistic styles? Start with something realistic, then make it painterly, then graphic-novel-like, then almost abstract.

That would be a cool way to compare how much the underlying composition survives when the visual language changes. It might also reveal which parts of the original concept are actually carrying the image.

[Brad92]

These are polished enough that I would be tempted to go in the opposite direction next time. Give us something strange. Weird architecture, impossible weather, a normal-looking room with one completely inexplicable object in it.

The photorealistic pieces are impressive, but a strong idea can stick in your head much longer than another technically perfect landscape.

[Fox50]

For portraits, I would experiment with expression and body language more than facial detail. A generated face can look convincing at first glance, but a person leaning against a wall, avoiding eye contact, fidgeting with an object, or reacting to something outside the frame can tell a much stronger story.

It also gives the image a reason to exist beyond showing off how realistic the face looks.

[Zero-Point]

The next set could be fun if you introduce scale as the main theme. Take an ordinary object and make it enormous, or make the environment gigantic while the subject is tiny.

A person standing beside an impossibly large greenhouse, a tiny train crossing a kitchen table, or a village built into the side of a teacup could all give you opportunities for dramatic composition without relying entirely on photorealism.

[Phoenix56]

Maybe try adding a little controlled ugliness. Harsh flash, blown highlights, motion blur, dirty windows, awkward perspectives, cheap fluorescent lighting, and slightly dull colors can all create character.

Perfect images are impressive, but recognizable visual flaws can make an image feel much more lived-in. Sometimes the glamorous lighting needs a break.

[Darren51]

The strongest images here seem to have a clear focal point, so I would keep pushing that. Before generating the next one, decide what you want the viewer to notice first, second, and third.

That sounds basic, but it is surprisingly easy for AI images to become collections of interesting details with no hierarchy. A strong photograph can contain lots of information while still making the eye land exactly where the photographer intended.

[CrimsonFury]

These are already a good reminder that AI imagery is much more interesting when there is some human direction behind it. The generator can produce endless variations, but deciding which variation is worth keeping is still a creative act.

For v2, I would perhaps make the project more cohesive rather than simply making the images more detailed. Pick a recurring idea, visual motif, location, or character and see how far you can take it. A smaller set with a clear identity could end up being much more memorable than a huge collection of unrelated beautiful pictures.

[Marcus82]

For the next portraits, accessories could become part of the storytelling. Instead of simply changing clothes, give the subject something meaningful to interact with: an old camera, a suitcase covered in stickers, a broken umbrella, a stack of letters, or a strange handmade object.

That gives the hands and posture something to do as well, which can make the scene feel less like a character standing still for a catalogue photo.

[ParallelSelf34]

Have you tried deliberately limiting the color palette? A scene built around two or three dominant colors can look much more cohesive than one where every surface is competing for attention.

For example, a cold blue interior with one warm orange light could give you a very different result from a scene where everything is equally saturated. It is a simple constraint, but constraints often make these experiments more interesting.

[veritas.io]

There is also something to be said for trying deliberately boring prompts. Generate an ordinary supermarket aisle, a bus stop on a cloudy afternoon, or somebody waiting for a train.

Those scenes are harder than they sound because there is nowhere to hide behind spectacle. If the result still feels like an interesting photograph, that says a lot about the consistency of the image generation.

[RayOfLight99]

The cinematic look is working, but I would watch out for making every image too cinematic. Once every picture has dramatic clouds, golden light, shallow depth of field, and a mysterious silhouette, the effect starts to cancel itself out.

Throwing in a few quiet images between the big ones could make the dramatic pieces hit harder. A calm frame can be a surprisingly good setup for the next spectacular one.

[FreeKickKing11]

A fun tangent would be to recreate the same fictional place across different decades. Make one image from the 1920s, another from the 1970s, one from the present, and one from the distant future.

Keep the architecture and geography consistent while changing clothing, vehicles, signage, technology, and atmosphere. If the location remains recognizable, you have basically created a tiny visual history of a place that never existed.

[PaleCipher]

Try some interior scenes next. AI tends to get a lot of attention for landscapes and portraits, but rooms can be fascinating because there are so many opportunities for repeated geometry, reflections, furniture, windows, and practical lighting.

A messy artist studio at 2 a.m. might actually be more interesting than another mountain vista. Give the model a complicated room and see whether it can maintain believable spatial relationships.