> ## Documentation Index
> Fetch the complete documentation index at: https://docs.apiyi.com/llms.txt
> Use this file to discover all available pages before exploring further.
# OpenAI gpt-image-2 Launches: Native 4K + 30% Price Cut
> OpenAI's flagship image model gpt-image-2 is here! Native support for any 2K/4K resolution, automatic high-fidelity reference images, 20-30% cheaper than 1.5. APIYI is live with zero-code OpenAI SDK migration.
## Key Highlights
* **🖼️ Native 2K/4K**: Single-shot output up to 3840×2160 (\~8.3MP), no upscaling pipeline needed
* **🎯 Auto High-Fidelity**: Reference image editing/fusion auto-enables high-fidelity, no manual `input_fidelity`
* **💰 20-30% Cheaper**: Token cost drops noticeably vs same-size, same-quality `gpt-image-1.5`
* **🌏 Native Chinese Prompts**: High-quality output without translation, stronger text rendering
* **🔌 Zero-Code OpenAI SDK**: Just point `base_url` to `api.apiyi.com/v1`
* **🛠️ Full Capabilities**: Text-to-image / reference editing / multi-image fusion (up to 5) / mask inpainting
## Background
In April 2026, OpenAI officially released **gpt-image-2**, the flagship upgrade to `gpt-image-1.5`. This is another structural step forward in OpenAI's image generation track: where 1.5 focused on "4x speed + precision editing," this generation tackles two long-standing pain points head-on — **resolution ceiling** and **per-unit cost**.
The most immediate change is **arbitrary valid sizes** — as long as you satisfy "max edge ≤ 3840px, both edges are multiples of 16, aspect ratio ≤ 3:1, total pixels 0.65MP–8.3MP," you can render directly. That means 4K landscape wallpapers, 1792×1024 cinematic frames, 3200×1800 infographics — all sizes that previously required upscaling post-processing — now come out in a single call.
The APIYI team integrated the model on day one. OpenAI's official SDK only needs a `base_url` change to call `gpt-image-2` — zero code migration.
## Detailed Analysis
### Core Features
Supports any valid output size. Presets cover 1K / 2K / 3840×2160 4K. Custom sizes only need to satisfy basic constraints (edges as multiples of 16, ratio ≤ 3:1).
Reference image editing automatically enables high-fidelity processing. Detail, character identity, and text retention dramatically improved. **Do not** pass `input_fidelity` — it will error.
1024×1024 high quality drops from the \$0.25 range of 1.5 to \$0.211/image. 2K/4K is token-metered but trends down equally — long-term cost noticeably lower.
Native Chinese prompt support. Stable rendering of Chinese/English text in signage, posters, UI screenshots. Fine text is rarely blurry on `high` quality.
### Performance & Specs
| Dimension | gpt-image-2 |
| ---------------------- | -------------------------------------------------- |
| Output resolution | Any valid size (1K/2K/4K, max 3840×2160) |
| Quality tiers | `auto` / `low` / `medium` / `high` |
| Output format | `png` (default) / `jpeg` / `webp` |
| Per call | 1 image (`n=1`) |
| Speed | \~120s (4K high quality approaches 2 min) |
| Chinese prompts | ✅ Native |
| Reference image limit | 5 (`image[]`) |
| Mask inpainting | ✅ Supported (alpha channel required) |
| Transparent background | ❌ Not supported (`background: transparent` errors) |
### Key Differences vs gpt-image-1.5
| Item | gpt-image-1.5 | **gpt-image-2** |
| ---------------------- | ----------------------- | --------------------------------- |
| Max resolution | 1024×1536 | **3840×2160 (4K)** |
| Custom sizes | Limited presets | **Any valid size** |
| Reference fidelity | Manual `input_fidelity` | **Auto-enabled** |
| Same-tier price | Baseline | **20-30% lower** |
| Transparent background | ✅ Supported | ❌ Not supported |
| Speed | \~30s | \~120s (trades for size/fidelity) |
Outputs above `2560×1440` (\~3.69MP) are officially marked **experimental** and may show quality fluctuations. For production, prefer presets like `2048x1152` / `2048x2048` / `3840x2160`.
## Real-World Applications
### Recommended Scenarios
Single-shot 4K (3840×2160 / 2160×3840). Perfect for movie posters, desktop wallpapers, video preview frames, large-screen materials — no upscaling pipeline needed.
Auto high-fidelity on reference images. Pass a character sheet to generate variations across scenes — identity, outfit, color palette retention significantly improved.
Up to 5 reference images + mask soft-guidance. Supports composite edit instructions like "subject from img1 + scene from img2 + style from img3."
Supports any aspect ratio within 3:1. 1792×1024 cinematic, 3200×1800 long-form, 2048×1152 video covers — all single-shot.
### Code Examples
#### Text-to-Image (Python, OpenAI SDK)
```python theme={null}
from openai import OpenAI
import base64
client = OpenAI(
api_key="your-apiyi-api-key",
base_url="https://api.apiyi.com/v1"
)
resp = client.images.generate(
model="gpt-image-2",
prompt="Cyberpunk city at night, neon sign closeup, cinematic frame",
size="2048x1152",
quality="high",
output_format="jpeg",
output_compression=85
)
with open("out.jpg", "wb") as f:
f.write(base64.b64decode(resp.data[0].b64_json))
```
#### Multi-Image Fusion + High-Fidelity Edit
```python theme={null}
resp = client.images.edit(
model="gpt-image-2",
image=[
open("person.png", "rb"),
open("scene.png", "rb"),
open("style.png", "rb"),
],
prompt="Place subject from img1 into scene from img2, using color style from img3",
size="1536x1024",
quality="high"
)
with open("edited.png", "wb") as f:
f.write(base64.b64decode(resp.data[0].b64_json))
```
**Response format**: `gpt-image-2` returns a **raw base64 string** (no `data:image/...;base64,` prefix). Decode it client-side to write a file, or prepend the prefix for browser rendering.
### Best Practices
**Production tips**:
* Prefer official presets (1024×1024 / 1536×1024 / 2048×1152 / 3840×2160) for stable speed/quality
* Default to `output_format=jpeg` + `output_compression=85` — faster than PNG, half the size
* Lock `quality=high` for text / signage / poster scenarios — lower tiers can still blur fine text
* Set client timeout **≥ 360 seconds** (conservative; `quality=high` + 2K/4K can take 3-5 minutes, and the \~120s figure causes many false timeouts)
* Exponential backoff on 5xx and timeouts, max 2 retries; log `x-request-id` for support
**Migration notes**:
* Code that passed `input_fidelity` must **remove the parameter** — the new model forces high-fidelity and will error
* `background: "transparent"` is unsupported — switch to `opaque` or post-process for transparency
* Still single-image per call (`n=1`) — issue parallel requests for multiple images
## Pricing & Availability
### Pricing (token-metered, common preset reference)
| Quality | 1024×1024 | 1024×1536 | 1536×1024 |
| ---------- | --------- | --------- | --------- |
| **Low** | \$0.006 | \$0.005 | \$0.005 |
| **Medium** | \$0.053 | \$0.041 | \$0.041 |
| **High** | \$0.211 | \$0.165 | \$0.165 |
**Pricing notes**:
* 2K/4K has no fixed per-image price — billed by actual input + output tokens
* Edit requests have noticeably higher input tokens than text-to-image due to forced high-fidelity
* Streaming (`stream: true` + `partial_images: N`) costs an extra 100 output image tokens per partial
* Source: OpenAI official pricing (April 2026)
### Stack with APIYI Recharge Promotions
On APIYI, beyond official token pricing, you can stack recharge bonuses for up to 20% additional discount. Details:
📖 Recharge promotions: `docs.apiyi.com/en/faq/recharge-promotions`
### gpt-image-2 vs gpt-image-2-all (Reverse)
| Pick | When |
| ----------------------------- | --------------------------------------------------------------------------- |
| **gpt-image-2** (Official) | Need precise size/quality control, depend on official API contract, want 4K |
| **gpt-image-2-all** (Reverse) | Want flat \$0.03/image, \~30s render, minimal parameters |
## Summary & Recommendations
`gpt-image-2` delivers "native large resolution + auto high-fidelity + same-tier price cut" all at once — a structural upgrade for **large-asset production** and **reference-driven editing** workflows.
### Recommended Use Cases
* ✅ **Design / Video teams**: Direct 4K posters, video covers, desktop wallpapers — skip the upscaling step
* ✅ **IP / character consistency**: Auto high-fidelity on references for character variations across scenes
* ✅ **Multi-image fusion workflows**: Up to 5 references + mask, composite edit instructions in one call
* ✅ **Smooth migration from 1.5**: Drop `input_fidelity`, avoid `transparent`, leave the rest as-is
### Usage Tips
1. **Stick with 1K presets when 4K isn't needed**: 1024×1024 / 1536×1024 are fastest and cheapest
2. **Budget extra for edit requests**: Forced high-fidelity means noticeably higher input tokens than pure text-to-image
3. **Timeout ≥ 360 seconds**: `quality=high` + 2K/4K can run 3-5 minutes — show progress in your UI
4. **Stick to presets for stability**: Sizes above 2560×1440 remain experimental — use cautiously in production
**Sources & dates**:
* OpenAI official docs: `developers.openai.com/api/docs/guides/image-generation`
* APIYI integration doc: `docs.apiyi.com/knowledge-base/gpt-image-2-API-for-user`
* Data accessed: April 23, 2026
***
Try `gpt-image-2` native 4K generation today — get an API key on APIYI and call directly from the OpenAI SDK with `base_url=https://api.apiyi.com/v1`!