Sudden severe drop in ChatGPT image quality and reference adherence

For consistency you can also add more parameters. Don’t be sparse on them. You can even tell about well established formats, like I mentioned Chicago, or Vancouver, can even try using some science oriented ISO ones. Or more publication related ones.

What I did notice is that a transparent background is a struggle for the model. My suggestion in this case is: use a contrasting unusual color as a background. And then ask ChatGPT to use python code to remove that color from the image. That makes a much cleaner output for now.

This is what I mean:

I used a Chroma background easy to remove

once removed you get a nice transparent transferable image to your document of choice.

Then in this case I applied a beige background:

This was a regen, not copy pasted to the end-document, just for concept explanation purposes.

YEAH I AGREE, FIRST SORRY FOR MY ENGLISH

BUT ITS SAD, THE NEW GPT IMAGE AFTER FINISHED THE IMAGE HE WILL SAID SOMETHING, BUT BACK VERSION JUST THE IMAVE WITHOUT TEXT

ALL MY WOMAN CHARACTER NOW GOT SENSITIVE LOL CAUSE SHE A WOMAN

I’VE ALREADY MAKE A THOUSAND WITH THE SAME PROMPT, PLS GPT BRING BACK THE OLD VERSION, I BUY GPT CAUSE FOR THIS AI IMAGE GENERATED pls

Pleass bring the last version before this, its a fail, l buy cause the ai image, but now my penny seems not worth it, after rendering image and finished, gpt now seems have a chat below the image, i know something different, my daily prompt cant used anymore, probably it can make me retiredd from game character design job, cause now all. My woman character seems get judge by her body, or outfit, is not nude or something, she still wearing, is a model fantasy, heuh…

Yes, I agree. I have been having the same problems. I have been off for a month and so haven’t generated images since end of June and when I tried today there was clearly a difference. What I got in June was so much better than now.

So the chicago and vancouver advice didnt help much since those are mainly academic citation styles, not visual/design formats. My work basically revolves around creating visuals-based images for kids to learn school subjects and topics.
Added more instructions, didnt help either, it just refuses to give me visually aesthetic images UNLESS I use ChatGPT Work.

Here’s the prompt I used:
"Create ONE portrait educational slide with a 9:16 aspect ratio (5.625 in × 10 in).

Audience and level:
Grade 9 Chemistry visual level only.
Do not write the grade on the slide.

Main goal:
Design a clean, polished, easy-to-read educational slide that explains and compares melting and boiling points in:

  1. metals
  2. simple molecular non-metals
  3. giant covalent substances

Title:
Melting and Boiling Points

Content:

Metals

Metals generally have high melting and boiling points.

A large amount of energy is often required to overcome the strong metallic bonding in their structures.

Simple molecular non-metals

Many simple molecular non-metals have relatively low melting and boiling points.

During melting or boiling, the relatively weak intermolecular forces between molecules are overcome.

The covalent bonds inside the molecules normally remain intact.

Giant covalent substances

Giant covalent substances have very high melting points.

A large amount of energy is required to break the strong covalent bonds extending throughout the structure.

These are general patterns and contain exceptions.

Design and layout instructions:

Use a clean three-part comparison layout with three clearly separated vertical panels or horizontal sections:

  • metallic structure
  • simple molecular structure
  • giant covalent structure

Place the title prominently at the top.

Below the title, arrange the three comparison sections in a balanced, visually even layout with strong alignment and consistent spacing.

For each section, include:

  • a clear section heading
  • a simple labelled structure diagram
  • a short text explanation using the supplied content only
  • a clear note showing what type of attraction is overcome during heating

Visual structure guidance:

For metals:

  • show a regular packed arrangement of metal particles
  • visually suggest strong metallic bonding
  • show that a large amount of energy is needed to overcome metallic bonding

For simple molecular non-metals:

  • show separate small molecules with space between them
  • clearly show weak intermolecular forces between molecules
  • clearly show covalent bonds inside each molecule
  • make it visually obvious that during melting or boiling, intermolecular forces are overcome
  • make it equally clear that the covalent bonds inside the molecules remain intact

For giant covalent substances:

  • show a continuous giant covalent network or lattice
  • visually suggest strong covalent bonds extending throughout the structure
  • show that a large amount of energy is needed to break these strong covalent bonds

Critical teaching clarity:
Clearly distinguish intermolecular forces from covalent bonds.
This distinction must be visually obvious, especially in the simple molecular non-metals section.

Use different visual treatments for:

  • intermolecular forces
  • covalent bonds
  • metallic bonding

For example, make them visibly distinct through line style, labeling, arrow style, or diagram treatment, so students can immediately tell them apart.

Do not state or imply that covalent bonds inside simple molecules are broken during boiling.

Text and typography:

  • Use concise, readable text blocks
  • Keep wording exactly aligned with the supplied content
  • Use strong visual hierarchy: title, section headings, body text, and labels
  • Make all text large enough to read easily on a phone screen or projected slide
  • Avoid dense paragraphs; break content into short chunks where needed
  • Use clear labels directly near the diagrams

Style:

  • clean educational infographic style
  • mature and neat, not childish
  • chemistry-focused and academically clear
  • minimal clutter
  • visually balanced
  • professional classroom-slide appearance

Colour and emphasis:

  • use a restrained academic colour palette
  • use colour coding only to improve comparison and clarity
  • keep the slide easy to read
  • ensure labels, arrows, and bonding/force distinctions stand out clearly

Composition:

  • keep generous margins
  • maintain even spacing between sections
  • avoid overcrowding
  • keep diagrams simple but accurate enough for Grade 9 understanding
  • make the comparison easy to scan at a glance

Important restrictions:
Use only the supplied content and requested visuals.
Do not add extra facts, examples, or decorative chemistry content.
Do not include quiz, test, or assessment elements.
Do not include mascots or cartoon characters."

This is the result I got. Its not about the image quality, its about the visual design. And my argument still stands. Why do Images 2.0 results vary based on whether you’re using Chat or Work? The second image, made in Work chat I used 5.6 Terra, light effort, standard speed. Still got a wayyy better result than first image made in regular Chat with 5.6 Sol High Intelligence.

Yeah for that age the examples I gave are too much. My point here was not that you adhere to my examples but to use template formats that are wide spread in the domain you intend to use them. Nine graders you can use a three card layout or a her visuals template. What is standard for the institution you work with?

Maybe this can help:

Slide-Layout Templates for Ninth-Grade Readers

Template Typical layout Works well for
Assertion–Evidence One complete-sentence claim above one strong visual Science, history, and persuasive lessons
Hero Visual Large image, short title, and minimal supporting text Introductions, themes, and emotional engagement
Split Screen Text or a question on one side, with a visual or example on the other Explanations and comparisons
Three-Card Layout Three equal blocks containing icons, examples, or steps Categories and concise summaries
Progressive Reveal One idea or stage introduced at a time Processes, arguments, and worked examples
Before–After Two contrasting panels Historical change, revision, and cause and effect
Comic Sequence Three to six sequential frames Scenarios, narratives, and social learning
Dashboard / Chapter Map A central topic surrounded by sections or progress markers Unit navigation and revision

A strong general formula is:

One idea + one visual anchor + one short explanation per slide.