Skip to main content
Use these models with the image generation and image editing endpoints. Reference a model by its id in the model field.
The exact set your key can call, and each model’s supported parameters, is returned by GET /v1/models. For per-model rates, see deepshi.ai.

Deepshi models

Deepshi’s own image models, running on Deepshi GPUs.

Deepshi Banana Pro

deepshi-banana-pro · Prompt 10KFast in-house text-to-image, tuned for quick, clean generations at any size. Dimensions

Deepshi Image 1

deepshi-image-1 · Prompt 10KIn-house model for text-to-image and multi-reference editing. Dimensions   Editing

Third-party models

Leading image models from across the industry, through the same endpoint, key, and balance. Recently added models appear first.

Kling Image O3

kling-image-o3 · Prompt 2.5KKuaishou’s flagship image model, generating and editing at up to 4K with multiple reference images. Resolution   Editing

Kling Image V3

kling-image-v3 · Prompt 2.5KA budget Kling tier at 1K and 2K, with single-image editing. Resolution   Editing

Reve 2.1

reve-2.1 · Prompt 4KReve’s model for exact prompt adherence and reliable in-image text, across a wide range of aspect ratios. Aspect ratio   Editing

Krea 2 Turbo

krea-2-turbo · Prompt 5KKrea’s fast text-to-image model with a polished, design-oriented look. Preset sizes

Ernie Image

ernie-image · Prompt 10KBaidu’s text-to-image model with eleven size presets and adjustable inference steps. Preset sizes

Cosmos 3 Super

cosmos-3-super · Prompt 4KNVIDIA’s model for physically grounded scenes with realistic lighting and geometry. Preset sizes

Qwen-Image 2

qwen-image-2 · Prompt 10KAlibaba’s model with strong text rendering, in English and Chinese. Preset sizes   Editing

HiDream-O1

hidream-o1 · Prompt 10KAn open model for generation, editing, and reference-based personalization. Preset sizes   Editing

Seedream 5 Pro

seedream-v5-pro · Prompt 10KByteDance’s flagship Seedream, strong at dense text and structured layouts. Preset sizes   Editing

FLUX.2 Turbo

flux-2-turbo · Prompt 10KThe fastest, lowest-cost tier of Black Forest Labs’ FLUX.2 family. Dimensions   Editing

FLUX.2 Flex

flux-2-flex · Prompt 10KFLUX.2 with adjustable steps and guidance, and up to ten reference images. Dimensions   Editing

Nano Banana 2 Lite

nano-banana-2-lite · Prompt 50KThe fast, low-cost tier of Nano Banana 2, at 1K. Aspect ratio   Editing

GPT Image 1 Mini

gpt-image-1-mini · Prompt 32KA compact, low-cost GPT Image tier with a selectable quality. Preset sizes   Editing

Nano Banana 2

nano-banana-2 · Prompt 50KGoogle’s fast, prompt-faithful generator with flexible aspect ratios, up to 4K. Resolution   Editing

GPT Image 1.5

gpt-image-1.5 · Prompt 32KOpenAI’s image model, with strong prompt adherence and text rendering. Preset sizes   Editing

Seedream 4.5

seedream-4.5 · Prompt 10KByteDance’s photorealistic generator with flexible preset sizes up to 4K. Preset sizes   Editing

Seedream 5 Lite

seedream-5-lite · Prompt 10KA lighter, faster Seedream variant for quick high-resolution generations. Preset sizes   Editing

Grok Imagine

grok-imagine · Prompt 8KxAI’s image model for fast, stylistic generations across many aspect ratios. Aspect ratio   Editing

Grok Imagine Quality

grok-imagine-quality · Prompt 8KThe higher-resolution Grok Imagine tier, for crisper detail at 2k. Resolution   Editing

Nano Banana Pro

nano-banana-pro · Prompt 50KThe pro tier of Nano Banana, for higher-fidelity detail and large 4K renders. Resolution   Editing

GPT Image 2

gpt-image-2 · Prompt 32KThe latest GPT Image model, with sharper detail and a selectable quality tier. Preset sizes   Editing

FLUX.2 Pro

flux-2-pro · Prompt 10KBlack Forest Labs’ high-fidelity generator with precise pixel-dimension control. Dimensions   Editing

Ideogram V4

ideogram-v4 · Prompt 10KIdeogram’s model, strong at typography and graphic design, with a selectable rendering speed. Dimensions   Editing
Not sure where to start? deepshi-image-1 is a capable in-house model for generation and editing, nano-banana-pro is a strong all-round default, gpt-image-2 is great for text in images, and seedream-4.5 is a fast, economical photorealistic option.