GPT Image 2 Now Powers Fleece AI's Image Tools (2026)
By Loïc Jané · Founder, Fleece AI
GPT Image 2 Now Powers Fleece AI's Image Tools
At a Glance: GPT Image 2 is OpenAI's image generation model unveiled on April 22, 2026. It supports 2K resolution, multilingual in-image text (English, French, Japanese, Korean, Chinese, Hindi, Bengali), high-fidelity inpainting and outpainting, and character/style consistency across edits — at $30 per million output tokens. Since September 2026, Fleece AI's
generate_imageandedit_imagetools run on GPT Image 2 (OpenAI model IDgpt-image-2, on the public API since April 21, 2026). See our AI workflow automation guide. Updated September 2026.
GPT Image 2 now powers Fleece AI's image tools (generate_image and edit_image) on every plan. When we first published this article in April 2026, our probe calls to the gpt-image-2 model ID were failing, and we announced the switch for early May 2026. It shipped later than that: the tools moved from GPT Image 1.5 to GPT Image 2 in September 2026. No action needed from you — existing agents and flows use the new engine on their next image call, with the same prompts.
Table of Contents
- What Is GPT Image 2?
- Key Takeaways
- Release Timeline and Availability
- What Will Change Versus GPT Image 1.5
- How to Use Image Generation on Fleece AI Today
- 7 Workflows That Will Get Better
- GPT Image 2 vs Other Image Models
- Pricing and Limits
- Frequently Asked Questions
Key Takeaways
- GPT Image 2 is an OpenAI image generation model unveiled April 22, 2026. It is live in ChatGPT, Codex and the developer API; OpenAI followed it with GPT Image 2.5 on September 8, 2026.
- The model supports up to 2K resolution (with a 4K beta), aspect ratios from 3:1 to 1:3, and multilingual in-image text rendering in English, French, Japanese, Korean, Chinese, Hindi, and Bengali.
- High-fidelity inpainting, outpainting, and character/style consistency across reference images make it the first OpenAI model genuinely usable for slide decks, infographics, product mockups, and marketing assets in a single call.
- Pricing on the OpenAI side is $30 per million output tokens and $8 per million image input tokens ($2 cached). On Fleece AI, image generation is included in your monthly credit allowance, with no separate billing.
- Since September 2026, Fleece AI's
generate_imageandedit_imagetools run on GPT Image 2 — every plan from Starter upward can use them. The switch came later than the early-May date we first announced. See our AI model comparison.
Release Timeline and Availability
The key dates:
- April 21–22, 2026 — GPT Image 2 launches in ChatGPT (web and mobile) and Codex, and the
gpt-image-2model ID (snapshotgpt-image-2-2026-04-21) is available on the public Images API (/v1/images/generationsand/v1/images/edits). - September 2026 — Fleece AI switches
generate_imageandedit_imagefrom GPT Image 1.5 to GPT Image 2, later than the early-May date we had announced. - September 8, 2026 — OpenAI releases GPT Image 2.5 in two variants,
gpt-image-2.5-flare(speed) andgpt-image-2.5-sunburst(editing precision), at the same token prices.
Fleece AI's image tools call the public API directly. No change to your prompts and no opt-in — your existing agents and flows use GPT Image 2 on their next image call.
What Is GPT Image 2?
GPT Image 2 is a frontier text-to-image and image-editing model developed by OpenAI. According to the official model documentation, the model will be exposed via the standard v1/images/generations and v1/images/edits endpoints under the model ID gpt-image-2, with a timestamped snapshot at gpt-image-2-2026-04-21.
OpenAI describes GPT Image 2 as the first image model to introduce O-series reasoning capabilities — the model can plan, search the web, and reason about composition before generating a frame. VentureBeat's launch coverage (April 2026) called the result "seemingly flawless" multilingual text rendering inside infographics, slides, and even manga panels.
For Fleece AI, this means a single tool call can now reliably produce things that previously required a designer: a French menu mockup with readable accents, a slide with three bullet points and a logo, an infographic with labeled axes, or a 2K product render that survives a press release.
What Will Change Versus GPT Image 1.5
Multilingual In-Image Text — Including French
GPT Image 1.5 could write English text inside images, but it stumbled on accents, ideograms, and any layout denser than a sign. GPT Image 2 ships native support for English, French, Japanese, Korean, Chinese, Hindi, and Bengali in-image text. For Fleece AI's French-speaking user base, this is the headline change — your agents can now generate Quebec menu boards, Parisian event posters, and bilingual product packaging that does not need a designer pass to fix typography.
2K Resolution and Flexible Aspect Ratios
The new model supports image sizes from 1024×1024 up to 2048×2048 in production (4K is in beta), with aspect ratios from a wide 3:1 to a tall 1:3. Fleece AI's generate_image and edit_image tools currently expose three sizes — 1024×1024, 1536×1024, and 1024×1536 — so the 2K sizes are available on the OpenAI API but not yet in Fleece AI.
True Inpainting and Outpainting in One Call
GPT Image 1.5 could "edit" images but really did remix-style regenerations. GPT Image 2 supports high-fidelity inpainting (modify a specific region while preserving the rest), outpainting (extend a canvas beyond the original frame), and large-scale image editing inside a single API call. On Fleece AI this maps directly to edit_image with source, style_reference, and composition_reference roles per attached image.
Character and Style Consistency Across Frames
The model can produce up to 8 images in a single prompt while preserving characters, objects, and style across frames. For Fleece AI agents that generate multi-panel content — a 4-step product walkthrough, a comic strip, a paginated infographic — this eliminates the previous need to re-prompt and pray.
Quality "auto" Mode
GPT Image 2 introduces an auto quality setting that lets the model pick low, medium, or high based on prompt complexity. Fleece AI's image tools expose the explicit low, medium (default), and high tiers.
How to Use Image Generation on Fleece AI Today
Fleece AI's image tools — powered by GPT Image 2 since September 2026 — are available on every plan, inside agents, flows, and chat.
To try image generation now:
- Open any agent (or create one) at fleeceai.app/dashboard/agents.
- Type "Generate a marketing poster for our spring sale, with the headline 'Spring -30%' over a golden landscape, 1536×1024 landscape format."
- The agent calls
generate_imageand within a few seconds returns a public URL embedded as a markdown image. - Reply "Now edit it: replace the headline with 'Spring -40%' and add small abstract decorations in the corners." The agent calls
edit_imagewith the previous URL as asourcereference.
Start now — open Fleece AI and ask any agent to generate or edit an image. New users get a 4-day Starter trial with full access to the image tools.
7 Workflows That Will Get Better
1. Multilingual Marketing Asset Generation
"Every Monday at 8 AM, generate three 1024×1024 social posts for this week's promotion — one in English, one in French, one in Spanish. Save them to Google Drive in the /marketing/weekly folder and post the French version to our LinkedIn page."
GPT Image 2's multilingual text rendering finally makes this a one-shot workflow instead of a designer ticket. Pair it with Google Drive automation for delivery.
2. Product Mockup From a Photo
"When I drop a product photo in this chat, generate a lifestyle mockup of the product on a marble countertop with morning light. Output 1536×1024 landscape."
edit_image with the photo as a source reference produces consistent renders across follow-up requests — the same product, different scenes.
3. Slide Deck Hero Images
"For each row in this Google Sheet of upcoming webinars, generate a 1536×1024 hero image with the webinar title rendered as text overlay, in our brand gold. Email me the URLs."
GPT Image 2 reliably types webinar titles inside the image — no separate text overlay step needed. Cross-link with Google Sheets automation.
4. Infographic Generation From Data
"Generate an infographic showing our Q1 2026 revenue: $2.4M total, +38% YoY, top 3 channels listed with percentages. Square format, brand colors gold and black."
The model's reasoning capability handles labeled charts and number rendering — previously a notorious weak spot for image models.
5. Style-Consistent Comic Panels
"Generate a 4-panel comic strip explaining what an AI agent does, in flat illustration style. Same character across all panels."
Up to 8 images per prompt with character consistency — finally usable for sequential storytelling.
6. Inpainting Edits at Scale
"Take the photo I just uploaded and replace the blue background with a sunset gradient. Keep the subject and lighting unchanged."
True inpainting means the subject is preserved pixel-for-pixel — not regenerated and almost-similar.
7. Localized Email Headers
"Every Tuesday morning, generate a 1536×1024 email header image for our newsletter with this week's headline rendered in French. Attach it to the draft Resend campaign."
Pair it with our transactional email patterns for a fully autonomous newsletter visual pipeline.
GPT Image 2 vs Other Image Models
| Feature | GPT Image 2 | GPT Image 1.5 | DALL·E 3 |
|---|---|---|---|
| Best For | All-purpose 2026 image work | Legacy GPT image flows | Older OpenAI integrations |
| Max Resolution | 2K (4K beta) | 1536×1024 | 1792×1024 |
| Multilingual In-Image Text | EN/FR/JA/KO/ZH/HI/BN | English (limited) | English only |
| Inpainting | High-fidelity, single call | Remix-style | Limited |
| Character Consistency | Up to 8 images | Single shot | Single shot |
| Reasoning Before Generation | Yes (O-series) | No | No |
| Quality Tiers | low/medium/high/auto | low/medium/high | standard/HD |
| Output Tokens (price) | $30 / M | n/a | n/a |
| Available on Fleece AI | Yes — all plans (since September 2026) | Replaced September 2026 | No |
GPT Image 2.5 (September 8, 2026) adds a speed-optimized Flare variant and a quality-optimized Sunburst variant at the same token prices as GPT Image 2; Fleece AI's tools run on GPT Image 2.
Pricing and Limits
According to OpenAI's image generation guide, GPT Image 2 charges $30 per million output tokens, $8 per million image input tokens, and $2 per million cached image tokens. On Fleece AI, every image generation or edit consumes credits from your monthly plan allowance — there is no separate per-image billing and no upgrade requirement.
Hard limits enforced by the model:
- Maximum edge length 3840 pixels.
- Both edges must be divisible by 16 pixels.
- Long-edge to short-edge ratio cannot exceed 3:1.
- Up to 8 images per prompt for character/style consistency runs.
On the OpenAI API, GPT Image 2 accepts sizes such as 2048x2048; Fleece AI's tools currently cap output at 1536×1024 or 1024×1536.
Frequently Asked Questions
How does GPT Image 2 compare to GPT Image 1.5?
GPT Image 2 is a strict upgrade: 2K resolution (vs 1.5K), multilingual in-image text in 7 languages (vs English-only), true inpainting and outpainting in a single call (vs remix-style edits), character consistency across up to 8 images, and a new "auto" quality tier. Fleece AI's tools switched from GPT Image 1.5 to GPT Image 2 in September 2026 — no work on your side.
Can GPT Image 2 integrate with Gmail, Slack, and Google Drive on Fleece AI?
Yes. Every generate_image and edit_image call returns a public HTTPS URL hosted on Vercel Blob, which any subsequent tool call can attach to a Gmail message, post to a Slack channel, or upload to Google Drive. See our Slack automation guide for a worked example.
Will GPT Image 2 be available on the Fleece AI Starter plan?
Yes. Image generation has always been a tool, not a model selector — every plan from Starter (€49/month) upward can call generate_image and edit_image inside agents, flows, and chat, and both run on GPT Image 2 with no plan upgrade required. See pricing.
How much does generating an image cost on Fleece AI?
Image generations are billed against your monthly credit allowance, not as separate charges. A single 1024×1024 medium-quality image typically consumes a small number of credits; a high-quality call consumes more. Starter (€49) includes 2,000 credits/month, Pro (€99) includes 8,000, and Business (€199) includes 20,000. See the full Fleece AI plan breakdown.
What are the known limitations of GPT Image 2?
Three honest caveats: (1) It is no longer OpenAI's newest image model — GPT Image 2.5 shipped on September 8, 2026 with faster (Flare) and more precise (Sunburst) variants. (2) Fleece AI's tools expose sizes up to 1536×1024 and the low/medium/high quality tiers; the 2K and 4K sizes and the auto tier are not exposed yet. (3) Multilingual text rendering is dramatically improved but not perfect for very dense layouts (e.g., a full magazine page in Bengali); for high-stakes typography, a human review pass is still recommended.
The Bottom Line
GPT Image 2 moves Fleece AI's image tools from "useful for placeholders" toward "production-ready for marketing, slides, and product visuals." Multilingual text, true inpainting, and character consistency across frames are the capabilities that made GPT Image 1.5 awkward to ship to customers. The switch arrived in September 2026, later than the May date we first announced, and your existing agents and flows use it without any change.
Open any agent and start building image workflows now.
Related Articles
- Claude Opus 4.7 on Fleece AI — Anthropic reasoning model on the Business plan
- GPT-5.2 on Fleece AI — tool calling, every plan
- Best AI Models for Workflow Automation 2026 — full model lineup
- What Is Fleece AI? — pillar overview
- AI Workflow Automation Complete Guide — pillar guide
- Best Autonomous AI Agents 2026 — platform roundup
- Automate Slack with AI Agents — post generated images to channels
- Automate Gmail with AI Agents — attach generated images to emails
Start building image workflows on Fleece AI — open any agent and ask for a picture today, generated with GPT Image 2.