Imagen 4 pricing and API access (2026): what it costs and how developers use it
Quick answer
Imagen 4 generation needs a migration plan, not just a price comparison. Google Cloud lists its three generation endpoints as discontinued June 30, 2026; the Gemini Developer API shut them down August 17. The $0.02/$0.04/$0.06 rates below remain published references, not proof of live access. Upscaling uses a separate preview endpoint. Check the provider’s current replacement before integrating. Rechecked October 10, 2026.
A quiet glass observatory above a cloud layer at blue hour, warm interior lights, one telescope, cinematic realistic lighting, balanced composition, fine architectural detail, no people, text, or logos.
Test an image brief before committing API spend
Generate one representative image, record the selected model and settings, and inspect text, faces, product details, and editability at full size.
Try AI Image GeneratorPublished Imagen 4 API rates and endpoint status
Google Cloud model or operation; verify availability | Published USD rate; not an access guarantee | Hypothetical 1,000 outputs |
|---|---|---|
$0.02 per image | $20 | |
$0.04 per image | $40 | |
$0.06 per image | $60 | |
$0.06 per image | $60 |
The totals are arithmetic examples before taxes, storage, other services, and additional generations. Upscaling is a separate operation. See Google Cloud's pricing table.
Can you still use the Imagen 4 API?
Through the Gemini Developer API: the standard, Ultra, and Fast Imagen 4 model IDs were shut down August 17, 2026. Google's deprecation table points to Gemini Nano Banana 2.1 (gemini-nano-banana-2.1) as the replacement. See the Gemini API deprecation schedule.
Through Google Cloud: Google's Imagen 4 model documentation lists imagen-4.0-fast-generate-001, imagen-4.0-generate-001 and imagen-4.0-ultra-generate-001 as discontinued June 30, 2026, with gemini-2.5-flash-image as the recommended Cloud migration. This notice covers those generation endpoints, not every Imagen product or third-party route.
How to estimate an actual project
This uses the published Imagen 4 rates to explain historical cost arithmetic; it is not a quote for a currently available generation endpoint. Multiply the number of charged generations by the model's per-image rate, then add any separate processing. For example, generating three Imagen 4 candidates for each of 100 accepted assets costs 300 × $0.04 = $12 in generation charges. Upscaling all 100 accepted assets at $0.06 each adds $6. The hypothetical total is $18 before other charges.
Compare accepted assets rather than raw images. A cheaper generation can cost more overall if it requires substantially more retries or editing.
Migration and integration steps
- Identify the provider and exact failing model ID. For the retired generation endpoints, select the replacement from that provider’s current migration notice—not an old SDK example.
- Check the replacement’s supported region, project access, billing and lifecycle. Retain the original prompt, approved reference files and output requirements.
- Adapt authentication, request structure and supported inputs using the replacement’s official example. Do not assume an Imagen parameter or upscaling operation transfers unchanged.
- Save the output and inspect text, composition, and intended subject details.
- Record charged attempts, errors, and accepted results before increasing volume.
Do not reuse an old Gemini API example after substituting only a hostname. Authentication, model IDs, request formats, and billing differ between providers.
Provider-specific migration matters. Google Cloud’s notice names Gemini 2.5 Flash Image; the Gemini Developer API’s current notice names Nano Banana 2.1. Cloud’s model lifecycle lists March 15, 2027 for Gemini 2.5 Flash Image. Read the schedule for the API you will actually call, rather than borrowing another provider’s retirement date.
Generation and upscaling are separate jobs. Google’s upscaling guide documents imagen-4.0-upscale-preview, with 2×, 3× and 4× factors and a 17-megapixel output ceiling. A 2048 × 2048 image at 2× becomes 4096 × 4096 (16.78 MP); at 4× it becomes 8192 × 8192 (67.11 MP), beyond that limit. Those are dimension calculations, not observed quality results or proof your project has preview access.
Standard, Ultra, or a different model?
Use the Standard/Ultra comparison for historical Imagen 4 work; do not start a new integration with the retired generation endpoints. For new generation or iterative editing, select a currently supported model with the inputs and edit operations your job requires. A higher published price does not establish better output. See our Imagen comparison and Nano Banana Pro pricing.
If you want to create an image without building an API integration, use a browser workflow such as Magic Hour's AI image generator. Its plans and available models are separate from Google's direct API rates.
Frequently asked questions
Through the Gemini Developer API: the standard, Ultra, and Fast Imagen 4 model IDs were shut down August 17, 2026. Google's deprecation table points to Gemini Nano Banana 2.1 (gemini-nano-banana-2.1) as the replacement. See the Gemini API deprecation schedule.
Through Google Cloud: Google's Imagen 4 model documentation lists imagen-4.0-fast-generate-001, imagen-4.0-generate-001 and imagen-4.0-ultra-generate-001 as discontinued June 30, 2026, with gemini-2.5-flash-image as the recommended Cloud migration. This notice covers those generation endpoints, not every Imagen product or third-party route.
Use the Standard/Ultra comparison for historical Imagen 4 work; do not start a new integration with the retired generation endpoints. For new generation or iterative editing, select a currently supported model with the inputs and edit operations your job requires. A higher published price does not establish better output. See our Imagen comparison and Nano Banana Pro pricing.
If you want to create an image without building an API integration, use a browser workflow such as Magic Hour's AI image generator. Its plans and available models are separate from Google's direct API rates.
