I passed back one of the prompts used to generate a prior image and it also got flagged.
@fofrai
-
DALL-E 3 Image Styles: Natural vs Vivid Quality Comparison
By
–
In dalle3, the real people live in the natural style. 1. vivid hd
2. vivid standard
3. natural hd
4. natural standard -
DALL-E 3 Settings Comparison: Vivid vs Natural Modes
By
–
I can't seem to keep prompt/seed consistent, but here's some comparisons of the 4 main dalle3 settings: 1. vivid standard
2. vivid hd
3. natural standard
4. natural hd -
Image Generation Style Parameters: Vivid vs Natural
By
–
There's one further param listed on the endpoint docs: – style, either "vivid" or "natural"
– vivid causes the model to lean towards generating hyper-real and dramatic images
– natural causes the model to produce more natural, less hyper-real looking images -
DALL-E 3 API: Key Features and Specifications Overview
By
–
Taking a look at the dalle3 API: – 1024×1024, 1024×1792 or 1792×1024
– 1 image per request (but can do more in parallel)
– prompts will be automatically re-written for safety, and to add more detail
– standard and HD quality, HD has more detail No seed, no negative prompts, no -
OpenAI Dev Day 2026 Keynote Livestream Announcement
By
–
OpenAI dev day keynote livestream here: https://
youtube.com/watch?v=U9mJuU
khUzk
… -
Better Illusion Model Released on Replicate Platform
By
–
Illusions may have had their moment, but I've been meaning to add this for a while: https://
replicate.com/fofr/illusions A better illusion model on Replicate with: – Deliberate v2 as base model
– img2img support
– inpainting/mask support
– controlnet start and end support
– resizing -
DALL-E 3 ChatGPT UI struggles with correction conflicts
By
–
The more I'm using dalle3 with ChatGPT, the more I'm fighting it. You correct one thing and it reverts another.
-
Halo Ring Fine-Tuning Delivers Promising AI Model Outputs
By
–
Getting some lovely outputs from the halo ring fine tune.
-
Fine-tuning SDXL for Upside Down Image Generation
By
–
SDXL isn't very good at upside down images. I wanted to play on this, so I fine-tuned the model on some upside down people (but didn't caption them as upside down). It extrapolates the upside-down-edness in a fascinating way. https://
replicate.com/fofr/sdxl-upsi
de-down
…