Skip to main content
Model selection belongs to each operation. Generate uses model_version; try-on, editing, inpainting, and video use model. A Studio interface label such as NB 2 is not the literal API value.

​
Start with these combinations

Kling 3 image-to-video also supports kling-v3-standard. The jobs API returns the same job shape for both models. The existing /images/{id}/video endpoint returns a parent image with a video version for kling-2.5. See jobs and versions.

​
Image quality and proportions

Send resolution explicitly as standard or high and start with num_images: 1. For Nano Banana 2 and Pro, Studio presents the quality choices as 2K and 4K. The output dimensions depend on the model and aspect ratio; read result.width and result.height from the completed job (width_px and height_px on the existing image endpoints). Generate and try-on default to one output when num_images is omitted. Edit and inpaint create exactly one output; use num_images: 1 or omit it. For Generate with Nano Banana 2 or Pro, common ratios include 1:1, 3:4, 4:5, 2:3, 9:16, 16:9, 4:3, 3:2, 5:4, and 21:9. The quickstart uses 3:4. Do not assume another model accepts every ratio. Generate and try-on support up to eight outputs where the model and account allow it. A larger batch, higher quality, or different model can use more credits or hit a different feature limit.

​
Outfit references

Generate with Nano Banana 2 or Pro can take multiple garment references, up to three. Choose one source form per product: outfit_asset_ids, outfit_image_ids, or outfit_image_urls. Keep alternative views of one garment together in an outfit asset. For Nano Banana 2 virtual try-on, use the singular outfit_asset_id, outfit_image_id, or outfit_image_url. The plural try-on fields are restricted to models that support them; a model being visible in Studio does not mean your API key has access to it. In particular, V4, V4x, V5, and V6 have additional API access restrictions.

​
Video inputs

Start with a single five-second shot. Use generate_audio to request audio on supported models, and multi_prompt for multiple Kling 3 shots. The shot durations determine the total duration and cost. Start and end frames must each be at least 300 × 300 pixels, no larger than 10 MiB, and have a width-to-height ratio between 0.4 and 2.5. When using both frames, their pixel dimensions must match. Do not supply both last_frame_id and last_frame_url.

​
Unsupported or restricted settings

Treat the HTTP error and its message as the next action: correct an invalid combination, choose an available model, or resolve an account limit. Do not silently switch models after a rejection if reproducibility or product appearance matters to your integration.