Skip to content

fix: change default model from chatgpt-image-latest to gpt-image-2#3

Open
Parkjoungwan wants to merge 1 commit into
openai:mainfrom
Parkjoungwan:fix/default-model-gpt-image-2
Open

fix: change default model from chatgpt-image-latest to gpt-image-2#3
Parkjoungwan wants to merge 1 commit into
openai:mainfrom
Parkjoungwan:fix/default-model-gpt-image-2

Conversation

@Parkjoungwan

@Parkjoungwan Parkjoungwan commented Jun 18, 2026

Copy link
Copy Markdown

Problem

chatgpt-image-latest no longer aliases to GPT Image 2. It currently points to an older model snapshot, producing noticeably lower-quality and less faithful style transfers.

Visual comparison — same photo, same prompt (Pathetic Scribble)

chatgpt-image-latest (current default) gpt-image-2 (this fix)
Over-polished illustration, color shading, realistic proportions Crude scribble drawn in an old computer painting program with a mouse — matches the original demo

The difference is stark: the current default produces what looks like a competent artistic sketch, while gpt-image-2 produces the intentionally terrible, clumsy output the prompt actually calls for.

Root cause

The chatgpt-image-latest alias is documented as "the image snapshot previously used in ChatGPT" — it pinned to an older version and no longer resolves to GPT Image 2. At the time the project was originally demonstrated, this alias did resolve to GPT Image 2. It no longer does.

Fix

Change the default model to gpt-image-2 directly.

# before
model: str = "chatgpt-image-latest",

# after
model: str = "gpt-image-2",

The existing input_fidelity guard already excludes gpt-image-2 and gpt-image-2-2026-04-21, so no further changes are needed.

Note on access

gpt-image-2 requires organization verification on platform.openai.com. This is consistent with the project's existing requirement for a paid OpenAI account.

chatgpt-image-latest no longer aliases to GPT Image 2 — it currently
points to an older snapshot, producing noticeably lower-quality and
less faithful style transfers (e.g. Pathetic Scribble, Goblin Mode).

Specifying gpt-image-2 directly restores the intended output quality
that the project was originally demonstrated with.

The existing input_fidelity guard already excludes gpt-image-2, so no
further changes are required.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant