laugh emoji!
hehe. Well done!
1 Like
Some comparisons…
I wish I would’ve kept better notes these last few years haha…
And I’m hungry now for some reason! lol
2 Likes
Another comparison…
An old owl muppet named Paul is wearing a yellow suit at the pondering pond in his old hometown, ultra-realistic fuzzy puppets
4 Likes
Had some issues with it reading from memory on 4o for the image directly, not entirely sure why yet…
Old Dall-E
Had to first read from memory
“Look at your memory for projects Phas, Phasm, Philosophical Ecosystems and PhasmIDs”
It printed out a summary
Then draw image in a separate message
“Draw a picture of this”
Then redrew with a better style formatting request
“Draw a graphically appealing ‘real life’ image of this”
5 Likes
D&D battlemap for a road encounter with ambush.. use 1" squares top down
then…
same map no people bushes on fire
4 Likes
I tried multi-layered capabilities.
This is the scenario to challenge:
- Text rendering
- Instruction following
- Diagram synthesis
- Multi-turn contextual consistency
Scenario: Emergency Response
Prompt 1:
A wide-angle, photorealistic view of a high-tech emergency response command room during a simulated multi-hazard drill. The room is filled with wall-mounted digital displays, illuminated control panels, and a round table in the center with three professionals (one firefighter in uniform, one woman in a public health jacket, and one man in a logistics vest) coordinating plans. The main display shows a detailed crisis map of a coastal city with icons for fire zones, flooding areas, and evacuation routes. One side screen shows a scrolling feed of emergency calls. Another screen displays a Gantt chart for resource deployment.
In the foreground, a digital whiteboard displays a hand-drawn “Emergency Response Protocol Flowchart” with arrows and boxes showing:
- “Hazard Detection” → “Threat Assessment” → “Response Activation”
- “Evacuation Coordination” ↔ “Shelter Allocation”
- “Medical Triage” → “Resource Replenishment” → “Recovery Planning”
Add realistic lighting from LED panels, reflections on screens, and legible text in a modern sans-serif font.
Prompt 2
Zoom into the digital whiteboard from the same scene. Make the flowchart visually clean, add icons (e.g., sirens, ambulance, water droplet, shelter symbol), and ensure the text is accurate and readable. The person closest to the board is pointing at the “Evacuation Coordination” node.
Prompt 3
Now generate a photorealistic handheld view from a first responder’s bodycam inside a flooded street of the same coastal city shown on the crisis map. Show ankle-deep water, emergency responders helping residents onto a truck, and a waterproof tablet displaying the same map used in the command room, now with “Zone 3 Evacuated” marked in red.
Prompt 4
Create a touchscreen dashboard close-up in the command room, showing a multilingual crisis status interface in Spanish and English. The display should include labeled buttons, a mini heatmap of affected zones, and live metrics. Buttons should include: “Activate Alarm / Activar Alarma,” “Send Alert / Enviar Alerta,” and “Update Map / Actualizar Mapa.” The UI must be sleek, responsive-looking, and all text must be legible in both languages, aligned correctly.
Prompt 5
Now show an alternative version of the coastal city map adapted for vision-impaired users: a tactile elevation model with raised textures and Braille labels. The surface is matte, slate-gray plastic, placed on a table with hands of an emergency worker gently tracing the raised roads. Include embossed symbols for shelters and danger zones. The Braille must follow standard positioning conventions.
Prompt 6
A dramatic, cinematic drone shot at dusk over the entire coastal city. We see fire zones glowing orange, floodwaters reflecting streetlights, traffic evacuating across bridges, and a helicopter with searchlights scanning rooftops. Label three zones faintly on the image: “Zone 1: Wildfire,” “Zone 2: Flooded,” and “Zone 3: Evacuated.” Use cinematic cool-warm lighting and depth haze to increase realism.
Prompt 7
A a warm, photorealistic scene set one week after the crisis in the same coastal city. The sun is shining. Community volunteers, emergency responders, and local families are working together in a cleanup and rebuilding effort. There’s a cleared street with fresh supplies being handed out. A banner across a relief tent reads “Together We Rise” in bold, clean font. In the background, children are drawing hopeful messages in colorful chalk. The firefighter and public health worker from the command room are seen smiling near the tent. Lighting is golden hour, with soft shadows and a calm, clear sky.
6 Likes
Sorry, not even the best picture ever, but this is my standard prompt across all the models over the years.
I threw in the humans to see how the new model renders them as well …
I love the realism in this new model myself.
5 Likes
5 Likes
Prompt: "Widescreen image. In the style of Rembrandt. Four puppies in medieval dress sit down to a beautiful dinner in a great castle’s main hall. Light streams through small windows, and the feast is bountiful. "
4o ImageGen:
DALL-E 3:
3 Likes
Image(Wide, Exact(Dawn Breaking Untouched Landscape))
Old
New
Just noted they changed the image sizes 1536x1024 rather than 1792x1024 or I have some issue with copying.
Cant find on intro page and model doesnt know
2 Likes
Your answer:
Obligatory content: I notice a strange effect here in the new image maker - the last pass that you finally receive as your output image can be significantly more distorted than the transmitted progress picture that you get a peek at.
You see two of three-wide and the top half, so we are a bit zoomed in on an attempt to do a multi-panel image, a replication of what I’ve done before with dalle at 4x4 that turned out pretty close to a series of photos.
She looks nice through just a bit of “filter” of not being done yet (to which I gave some color balance and sharpening), then - whoa, a monster!
2 Likes
Vote for the best images!
Since this thread is all about sharing the best images, I encourage everyone to like their favorites.
The most liked images will be featured in the top post, along with proper credit to the creators and a nod to their skills.
Marking this call to action as the solution to give it extra visibility.
Now, let’s see how far you can push this model!
1 Like
This are the most interesting ones, to see model changes…
3 Likes
“A hipster Owl animator named Heidi paints a self-portrait of herself with her eyes open. Hedi’s face is wise like an owl with eyes like stars,”
no you did it wrong… she should be normal but the image of herself she paints is exaggerated and goofy
4 Likes
side by side, using the same prompton Dall-e and 4o:
DALL-E (less than 10 seconds)
GPT 4.5 (1 minute and half)
Side by side again
Dall-E
GPT 4.5
Dall-E
(i asked a landscape
)
Now same but this time I have not used prompt of dall-E. I asked him to use the image, and reproduce it better.
Better…
worse for characters, better for constructions and light
far better
Conclusion:
The new generator seems interesting. It is however really slow, and by far compared to dall-e (a little too much for me… more than 1 minute for 1 image, that’s huge).
It reacts really badly to the prompt and images lack of depth, clearly.
But, it may have a potential
But what is really troubling is the ability to reproduce images… Look the results when I asked him to reproduce images…
3 Likes
