GPT Image 2
OpenAI · AI imageGPT Image 2 is OpenAI's image generation and editing model, available on Replicate for API-style access. It produces photorealistic scenes with strong materials and lighting, renders dense text and complex layouts (infographics, UI mockups, marketing), and supports precise edits that preserve identity, composition, and lighting when you supply reference images.
On VidMachine, GPT Image 2 is offered as an image option for start frames and AI Images projects. We always call it at medium quality for consistent speed and cost. The Replicate-hosted model exposes aspect ratios 1:1, 3:2, and 2:3—not native 9:16—so when you need vertical shorts (9:16), VidMachine requests a 2:3 image and then center-crops the sides so the saved asset matches a true 9:16 frame without changing height.
Each generated image costs 10 credits on VidMachine. For transparent PNG backgrounds, OpenAI documents other model paths; GPT Image 2 does not support transparent backgrounds on Replicate.
Key features and benefits
Generation and editing
GPT Image 2 handles text-to-image generation and image editing with instruction following. Pass one reference image for targeted edits or multiple images to combine styles, subjects, or references into one output. The model aims to preserve high fidelity on inputs automatically.
Photorealism and text
Expect natural lighting, believable materials, and readable text when you quote exact copy and describe typography. The model is suited to thumbnails, hero frames, infographics, and branded shorts where legibility matters.
Aspect ratios on VidMachine
The upstream API supports 1:1, 3:2, and 2:3. For 9:16 vertical shorts, VidMachine generates at 2:3 (closest portrait preset), then crops the left and right evenly so the result is centered 9:16 at the same pixel height—matching how Shorts and TikTok frames are used downstream.
Quality tier
Replicate exposes quality settings including low, medium, high, and auto. VidMachine always uses medium quality for GPT Image 2 so billing and behavior stay predictable.
Technical specifications
HostReplicate (openai/gpt-image-2)
Aspect ratios (API)1:1, 3:2, 2:3
VidMachine 9:16Generate 2:3, then center-crop to 9:16
Quality on VidMachineMedium (fixed)
Credits per image10
Use cases and applications
GPT Image 2 fits creators who want OpenAI-quality imagery with editing and multi-reference workflows without leaving VidMachine. Use it for polished start frames, story-driven thumbnails, and layouts that include on-image text.
Pair it with your video model priority when you care about prompt adherence and edits from reference images. Remember that 9:16 outputs are derived from 2:3 plus a centered crop—compose important subjects near the vertical center when possible.
Why this model
Choose GPT Image 2 when you want OpenAI's latest general-purpose image stack at a fixed medium-quality tier and a clear 10-credit price per image. It competes with other premium image options on VidMachine while exposing familiar OpenAI strengths in text and editing.
If you need native 9:16 without cropping, pick another image model that supports that aspect ratio directly. If you need transparent PNGs, use a model that supports transparency per provider docs.
How VidMachine uses it
Add GPT Image 2 to your project's image model priority. Scene and cover generation use your selected image model; credits deduct per image at the rate above.
For portrait shorts, VidMachine maps 9:16 to a 2:3 API request and post-processes the file before upload to storage so downstream video and layouts see true 9:16 dimensions.
What you should know
Why does VidMachine crop for 9:16?
GPT Image 2 on Replicate does not offer a 9:16 preset. We use 2:3 (tall portrait) and center-crop horizontally to 9:16 so the height matches the generated image and the frame fits vertical video.
How many credits is each image?
Each GPT Image 2 image costs 10 credits on VidMachine.
Can I change quality to high or low?
VidMachine always uses medium quality for this model so runs stay consistent and predictable.