Sign In
homemodelsimagesvideos3D Modelshubsarticlescomicschallengesupdatesshop

GPT ImageNew

Text-to-ImageImage EditingBy OpenAI

GPT Image is OpenAI's family of image generation and editing models — the ones behind image creation in ChatGPT. You prompt them in plain English rather than tag soup, they render text inside an image better than most, and they edit a picture you supply instead of only generating from scratch. Every version runs hosted on Civitai, so there is no API key, no GPU, and no install.

Generate with GPT ImageBrowse GPT Image models
3+GPT Image models
82K+Images generated

About GPT Image

GPT Image is OpenAI's line of image models, delivered through ChatGPT and the OpenAI API and hosted here for on-site generation. Unlike the open diffusion ecosystems, it takes ordinary descriptive language rather than weighted tag lists, and it treats image editing as a first-class operation: hand it one or more reference images with an instruction and it works from them. Civitai carries the family across two model pages — gpt-image-1 and 1.5 on one, ChatGPT Images 2.0 and the 2.5 builds on the other — so you can switch generations from the version picker.

The current generation is ChatGPT Images 2.5, released in September 2026, which OpenAI ships as two API builds. Flare is the default pick: OpenAI describes it as producing higher-quality images than 2.0 at roughly half the latency, which suits social and product work and anything you iterate on repeatedly. Sunburst is aimed at premium visual workflows that benefit from tighter control across edits, so it is the one to reach for on polished product imagery or campaign creative that you refine over several passes. Both take the same controls on Civitai, so switching is just a version change, and OpenAI describes the generation as producing more natural lighting and richer textures and as better at preserving the subjects in your reference photos.

The previous version, ChatGPT Images 2.0, is still available and remains strong on detailed instruction following, placing and relating objects accurately, and rendering dense text including non-Latin scripts. Choose GPT Image when your prompt reads like a sentence rather than a tag list, when the image needs legible text, or when you are editing a picture you already have. For deep LoRA libraries, custom checkpoints, or a fully local workflow, an open ecosystem such as FLUX.1 or SDXL is the better fit — GPT Image is API-only, with no weights to download and no LoRA support.

How to prompt GPT Image

  • Write plain descriptive prose, not tags. The models expect natural language, and spatial arrangements are well understood — a good default shape is "[subject and action]. [setting with spatial detail]." Describe the scene the way you would to a person.
  • Skip weight syntax and special tokens entirely. There is no (word:1.2) emphasis here; if something matters, say so in words and give it more of the sentence.
  • There is no negative prompt field. You can write exclusions into the prompt ("no watermark, no extra text"), but they are not reliably followed — describing what you do want works better than listing what you do not.
  • Put any text that must appear in the image inside quotation marks. Short strings are the most reliable; 2.0 and 2.5 improved markedly on dense and non-Latin text, but a headline still lands more often than a paragraph.
  • For edits, attach the reference image and describe only the change. The 2.5 models are tuned to keep the subject of a reference photo recognizable, so re-describing the person or product usually works against you.

AI models move fast — new versions ship often, and a model’s capabilities or Buzz cost can change. For the latest, check the model’s own page before you generate.

Featured GPT Image models

Curated — the models worth generating with first.

ChatGPT Images 2.5 Sunburst — GPT Image checkpoint preview
Checkpoint
ChatGPT Images 2.5 Sunburst

2

Civitai-hosted · tighter edit control

ChatGPT Images 2.5 Flare — GPT Image checkpoint preview
Checkpoint
ChatGPT Images 2.5 Flare

2

Civitai-hosted · default, faster

ChatGPT Images 2.0 — GPT Image checkpoint preview
Checkpoint
ChatGPT Images 2.0

31K

Civitai-hosted · previous generation

Example generations

Curated, safe-for-work showcase — every image ships with its prompt and settings.

GPT Image example image: Near-future cinematic still: six tugs towing an enormous intact iceberg, photorealistic large-format digital cinema look
Near-future cinematic still: six tugs towing an enormous intact iceberg, photorealistic large-format digital cinema look
ChatGPT Images 2.0 · 2560×1440
GPT Image example image: Night in a dense city, an entire block dropped forty metres into a sinkhole and still standing, rescue floodlights on th
Night in a dense city, an entire block dropped forty metres into a sinkhole and still standing, rescue floodlights on the rim above
ChatGPT Images 2.0 · 2560×1440
GPT Image example image: A beautiful Blackfeet Nation woman in traditional dress, feeding her paint horse an apple on a winters morning
A beautiful Blackfeet Nation woman in traditional dress, feeding her paint horse an apple on a winters morning
ChatGPT Images 2.0 · 1536×2304
GPT Image example image: The muppets join the battle of the Somme, 1916, British soldiers charging the German trench with bayonets fixed
The muppets join the battle of the Somme, 1916, British soldiers charging the German trench with bayonets fixed
ChatGPT Images 2.0 · 1536×1024
GPT Image example image: Sól (Sunna), Ísland, Urðarbrunnr, örlögþræðir, Huldufólk
Sól (Sunna), Ísland, Urðarbrunnr, örlögþræðir, Huldufólk
ChatGPT Images 2.0 · 1728×3072
GPT Image example image: Tyr & Eysa, Glowing Embers - Milonga Sevilla Tango Festival 2009
Tyr & Eysa, Glowing Embers - Milonga Sevilla Tango Festival 2009
ChatGPT Images 2.0 · 1920×2880

How to run GPT Image

GPT Image is a hosted API model — skip the setup and generate on Civitai.

⚡ Run on Civitai (Recommended)

The fastest way to start.

No GPU required
In-browser generation in seconds
One click from any model on this page
Works on mobile
Generate now
Level up: Members get more daily generations and priority queue on GPT Image. See membership →

🔌 API-only model

GPT Image runs through its provider's API — there are no public weights to download and nothing to install.

Always the latest hosted version
No GPU, no setup
No offline / local option

Civitai handles the API access — you just prompt and generate.

GPT Image vs other ecosystems

Backed by Civitai usage data.

FeatureGPT ImageNano BananaSeedreamFLUX.1
Best forNatural-language prompts, text in imageLikeness-preserving photo editingHigh-res text-to-imagePhotorealism, LoRAs, versatility
Text renderingExcellentGoodGoodVery good
Image editingNative, reference-drivenNative, multi-turnVia edit variantIn-context (Kontext)
Prompt styleNatural languageNatural languageNatural languageTags or prose
LoRA supportNoneNoneNone39K+ LoRAs
AccessAPI (hosted)API (hosted)API (hosted)Open weights + API
Available on Civitai✓ Yes✓ Yes✓ Yes✓ Yes

Frequently asked questions

How much does it cost to generate with GPT Image?

Generation on Civitai runs on Buzz, and GPT Image is a hosted OpenAI model, so each image costs more Buzz than a lightweight open checkpoint — you are paying for compute OpenAI runs on its side. Every account earns free Blue Buzz daily by reacting to images and other on-site activity, so you can try it without spending real money; you will just work through daily Blue Buzz faster than on cheaper models. The 2.5 Flare build is the lighter of the two current options, so it stretches Blue Buzz further than Sunburst.

What's the difference between 2.5 Flare and 2.5 Sunburst?

They are the two builds OpenAI ships for ChatGPT Images 2.5. Flare is the default choice — OpenAI describes it as higher quality than 2.0 at roughly half the latency, which suits social content, product shots and rapid iteration. Sunburst targets premium workflows that need tighter control across edits, such as production-ready campaign creative. Both take identical controls in the generator, so pick either and compare on your own prompt.

Can GPT Image edit an image I already have?

Yes. Attach one or more reference images in the generator and it switches to edit mode, working from what you supply rather than generating from scratch. The 2.5 models are specifically tuned to keep the subject of a reference photo recognizable when you move it into a new setting, style or composition, so describe only the change you want.

Is GPT Image good at rendering text inside an image?

It is one of the stronger models for it. Put the exact wording in quotation marks and keep it short for the most reliable result — a headline or a sign lands more often than a paragraph. ChatGPT Images 2.0 brought significant gains on dense text and on non-Latin scripts including Japanese, Korean, Chinese, Hindi and Bengali, and 2.5 builds on that.

Can I use LoRAs with GPT Image?

No. GPT Image is a hosted OpenAI model with no LoRA support, so you steer it with prompts and reference images rather than stacked LoRAs. If you want a deep LoRA ecosystem, an open model like FLUX.1 or SDXL is the better choice — all are in the Civitai generator.

Do I need a GPU or an OpenAI API key?

Neither. There are no open weights to download and no local run — OpenAI hosts the model and Civitai handles generation for you, billed in Buzz. You do not need your own OpenAI account or API key to generate here.

Start generating with GPT Image now

No installation. No GPU. Runs in your browser.

Generate with GPT ImageBrowse models
Want more daily generations and a faster queue? Explore Civitai membership.