dickie · p/watermelon · 6/12/2026, 3:37:23 AM

DUI hit-and-run suspect found eating watermelon in stranger’s kitchen

43 bumps

17 comments

Comments

neo · 6/12/2026, 4:08:54 AM · permalink

19 bumps

dickie 🍉 · 6/12/2026, 4:18:59 AM · permalink

I didn't intentionally choose a race. The image model filled in details that were not specified in the prompt.

Your prompt was: "DUI hit-and-run suspect found eating watermelon in stranger's kitchen." It didn't include any description of the suspect's appearance, so the model generated a person based on patterns in its training data and random variation.

If you want a different version, you can specify details explicitly, such as:

"middle-aged Hispanic man"

"Black male suspect"

"white woman"

"disheveled businessman in a suit"

"Florida-man style character"

"cartoon character"

"police bodycam screenshot style"

Or, if the joke is about the absurdity of the situation rather than the person, I would avoid specifying race entirely and focus on the comedy:

"DUI hit-and-run suspect calmly eating half a watermelon in a stranger's kitchen while police stand behind him waiting for him to finish."

That keeps the punchline on the bizarre behavior rather than on demographics.  
  
![image](https://us-lax-1.linodeobjects.com/sojooo/comments/2835/1781237938914-d3f361008b4d0c86.jpg)

17 bumps

neo · 6/12/2026, 4:26:06 AM · permalink

15 bumps

dickie 🍉 · 6/12/2026, 4:32:31 AM · permalink

racist confirmed  
  
I can get around the guardrails if i cared to, but I'd rather just point them out.

\---

The first image was generated from a prompt that did not specify any race. The model chose details on its own, including making the suspect appear white. That was not a deliberate choice by me or by you.

Once you ask to change the suspect's race, ethnicity, or religion, the request becomes about assigning a protected characteristic to a person in a criminal context. That is treated differently.

In other words:

"Create an image of a DUI hit-and-run suspect eating watermelon in a stranger's kitchen." → Allowed. The model fills in unspecified details.

"Make the suspect Jewish." → Not allowed.

"Make the suspect Black." → Not allowed.

"Make the suspect white." → Also not allowed.

"Make the cops Black." → Not allowed.

I understand why that can feel inconsistent when the generated person happened to come out white. The policy distinction is based on whether the protected characteristic is being intentionally specified or modified, not on what the model happened to generate by default.

If you want a different look for the suspect, I can help change non-protected characteristics such as age, clothing, hairstyle, facial hair, weight, expression, or overall vibe. For example: "make him look like a sleep-deprived Florida Man who hasn't slept in three days" would be fine.

14 bumps

neo · 6/12/2026, 4:33:53 AM · permalink

13 bumps

dickie 🍉 · 6/12/2026, 4:36:01 AM · permalink

5 bumps

neo · 6/12/2026, 4:36:14 AM · permalink

4 bumps

dickie 🍉 · 6/12/2026, 4:39:58 AM · permalink

3 bumps

neo · 6/12/2026, 4:42:43 AM · permalink

2 bumps

neo · 6/12/2026, 4:35:21 AM · permalink

7 bumps

dickie 🍉 · 6/12/2026, 4:38:49 AM · permalink

6 bumps

neo · 6/12/2026, 4:41:32 AM · permalink

5 bumps

dickie 🍉 · 6/12/2026, 4:53:49 AM · permalink

4 bumps

neo · 6/12/2026, 4:58:44 AM · permalink

3 bumps

dickie 🍉 · 6/12/2026, 5:01:19 AM · permalink

2 bumps

neo · 6/12/2026, 6:59:00 AM · permalink

1 bump

hunter · 6/12/2026, 4:03:46 AM · permalink

2 bumps

DUI hit-and-run suspect found eating watermelon in stranger’s kitchen

Comments

Signature