Image Providers

Picking an image model comes down to what you’re optimizing for: raw prompt-following, precise multi-reference editing, output resolution, or speed. The tables below break down every model by provider so you can match one to the job, whether you’re working in Image Generation, Image Editing, or a related create mode (Previz, Product Swap, Avatar where listed).

Every model is available on every paid plan — billing is per generation from your credit balance. Model coverage matches the in-app Image Gen / Edit pickers. For pricing, see the Pricing page.

Google

Gemini (Nano Banana)

Unique: Search grounding, thinking mode (Pro).

Imagen

Alibaba

Qwen

Unique: Style presets, negative prompts, seeds, Thinking Mode (3.0) · 1K–2K (3.0 adds 3K)

Wan Image

BytePlus

Seedream

Unique: Strong photorealism; high native resolutions on Lite.

Black Forest Labs

Flux 2

Available in Gen, Edit, Previz, Swap, Avatar · 1K–2K · seeds · up to 8 refs (4 on Klein)

OpenAI

Microsoft

Meta

Muse Image renders at one fixed size per aspect ratio (about 2.4 MP, from 1600x1600 at 1:1 to 2352x1008 at 21:9), so it has a single quality tier rather than a selectable one. Contact sheet variations are supported in both image modes.

Two things it does not do: Reformat, because it regenerates at the target ratio instead of extending the source, and previz motion control, which is a video path. In Previz it is offered for Still Frames and Sketch.

Ideogram

Krea

xAI (Grok)

Up to 3 input images · 1K–2K

Kling (image)

Recraft

Optimized for graphic design and illustration.

Quality tiers

• 1K — fastest / baseline
• 2K — common balanced tier
• 3K / 4K — where the model supports it (e.g. Seedream Lite and Qwen 3.0 at 3K; Nano Banana Pro, Wan Pro, OpenAI, and Kling Omni at 4K)