What can GPT Image 2.0 generate on Soup.video?

Text-to-image when the tray is empty. Image-edit when you attach stills — style transfer, object changes, compositing, and character-consistent edits.

How do I use GPT Image 2.0?

Write a prompt. Attach stills or skip them. Set ratio and quality (High by default). Cook. Closing the tab after cook starts is fine; Soup.video emails you when the still is ready.

How does text-to-image work?

Leave the image tray empty. The prompt becomes the whole frame. Set ratio and quality, then cook.

How does image-edit work?

Attach PNG, JPEG, or WebP stills. The prompt describes the change — restyle, combine, swap a background, keep a character. The model preserves identity, composition, and lighting unless you ask otherwise.

What quality tiers and prices does GPT Image 2.0 support?

Low is 2 credits per image, Medium is 6, High is 15, Auto is 15. Same price for generate and edit. High is the default.

GPT Image 2.0 AI Image Generator

Studio review wall of printed AI image sheets with sticky notes — Candid photograph of a creative studio's review wall covered in a neat grid of printed image sheets and loose pinned photographs showing a range of AI generated visuals: product shots, portraits, landscapes, abstract compositions. Some prints circled in red grease pencil, a few with handwritten sticky notes saying "RERUN", "FINAL", and "GPT-IMAGE-2".

Corporate boardroom meeting with a rising revenue chart on a TV — Photorealistic corporate boardroom: a long polished mahogany conference table with leather executive chairs, a glass of water at each place setting, notepads and pens arranged neatly, a large flat-screen TV at the end showing a quarterly revenue chart going up. Floor-to-ceiling windows with a city skyline view. Serious, sterile corporate atmosphere. Several business executives in suits seated around the table in a meeting.

Vintage vinyl album sleeve for Distances by Clara Veil — Transform this into a vinyl album cover. Add a bold album title "DISTANCES" in a clean condensed sans-serif across the top. Artist name "CLARA VEIL" in smaller type below it. A parental advisory sticker in the bottom right corner. The image gains a slight halftone print texture and the colors shift toward faded vintage tones. Add a thin white border around the entire image like a physical print. A small record label logo in the bottom left corner reading "MERIDIAN RECORDS". It should look like you're holding an actual vinyl sleeve.

Double-decker bus wrapped in full branding livery on a city street — Wrap the double-decker bus from the first image with the exact branding design from the second image as a full bus livery. Keep the street scene, people, and background unchanged.

Ceramic Soup.video bowl on a creative studio desk — Transform the attached Soup.video logo into a real-world physical object: a playful, premium ceramic soup bowl sitting on a busy creative studio desk, with the Soup.video logo naturally printed/embossed on the front of the bowl. The bowl contains glowing neon-orange soup with tiny floating film-strip and camera-shaped ingredients. Around it are a cinema camera, storyboard sketches, headphones, scattered sticky notes and a laptop showing colorful AI-generated video frames. Warm cinematic studio lighting, realistic materials, subtle reflections, shallow depth of field, high-end commercial photography, slightly quirky and fun, photorealistic. Preserve the attached Soup.video logo exactly — do not redesign, distort, misspell, or reinterpret the logo. Make the logo clearly visible and naturally integrated into the physical object. 4:5 composition.

Why GPT Image 2.0?

Text to image

Describe a scene and get a high-fidelity still.

Precise edit

Pass one or more stills. Change only what you ask for.

Readable text

Dense lettering, UI mockups, infographics, and logos.

Quality tiers

Low, Medium, High, or Auto. High is the default.

Character lock

Keep the same face across stills with reference images.

Pricing

Credits per image

Quality	Credits
low	✦ 2
medium	✦ 6
high	✦ 15
auto	✦ 15

How GPT Image 2.0 works

GPT Image 2.0 is OpenAI’s latest image model. It generates stills from text and edits existing images with close instruction following — photoreal detail, sharp in-image type, and targeted changes that leave the rest of the frame alone.

Quality is Low, Medium, High, or Auto. High is the default. Attach PNG, JPEG, or WebP. Pass one still to edit it, or several to combine style, subject, and reference into one output. Transparent backgrounds are not supported.

Capabilities

text-to-image — 0 stills. The prompt becomes the whole frame.
image-edit — 1 or more stills. Transform, restyle, compose, or keep a character while you rewrite the scene.

How to use

Everything happens in the cook bar at the bottom of the page.

1. Prompt — describe the still, or the edit you want on attached images.
2. Stills — skip them for generate, or attach references for edit / compose / character lock.
3. Settings — ratio (auto, 1:1, 3:2, 2:3) and quality (Low / Medium / High / Auto).
4. Cook — the button shows the quality price. Closing the tab after cook starts is fine; we email you when the still is ready.

Text to image

Generate a still from a written description — photoreal, illustration, product, infographic, or readable in-image type. Put exact copy in "quotes" and describe the typography.

How — leave the image tray empty. Write the prompt, set ratio / quality, cook.
Best for original scenes, logos, comics, UI mockups, and marketing visuals from text alone.

Image edit

Transform existing stills with a prompt. One image is a directed edit. Several images fuse style, character, and layout into one frame. Say what must stay the same.

How — attach PNG, JPEG, or WebP stills. Describe the change, then cook. Label references by number when you pass more than one.
Best for restyles, product mockups, lighting changes, object removal, and character-consistent series.

Settings

prompt: Required. Describe the image to generate, or the edit to apply.
images: Optional. Skip for text-to-image. Attach stills (PNG, JPEG, WebP) to edit, compose, or use as character/style reference.
aspect_ratio: auto (default), 1:1 square, 3:2 landscape, or 2:3 portrait.
quality: low, medium, high (default), or auto. Lower is faster and cheaper. Auto is billed as High.

About

GPT Image 2 is OpenAI’s state-of-the-art image generation and editing model. It is served here via Replicate as openai/gpt-image-2.

FAQ

What can GPT Image 2.0 generate on Soup.video?: Text-to-image when the tray is empty. Image-edit when you attach stills — style transfer, object changes, compositing, and character-consistent edits.
How do I use GPT Image 2.0?: Write a prompt. Attach stills or skip them. Set ratio and quality (High by default). Cook. Closing the tab after cook starts is fine; Soup.video emails you when the still is ready.
How does text-to-image work?: Leave the image tray empty. The prompt becomes the whole frame. Set ratio and quality, then cook.
How does image-edit work?: Attach PNG, JPEG, or WebP stills. The prompt describes the change — restyle, combine, swap a background, keep a character. The model preserves identity, composition, and lighting unless you ask otherwise.
What quality tiers and prices does GPT Image 2.0 support?: Low is ✦ 2 per image, Medium is ✦ 6, High is ✦ 15, Auto is ✦ 15. Same price for generate and edit. High is the default.