Ctrl K
Media production lab

Reference Stills

Compare GPT Image 2 and Nano Banana while preparing reliable video anchors.

Updated: September 2026

Know the canvas labels

Image Generation uses 302.AI. Nano Banana Pro uses gemini-3-pro-image-preview, including when an older canvas node still stores a retired Flash or TokenStar alias.

  1. Place Image → Image Generation, select it, then open the bottom Model menu to inspect GPT Image 2 and Nano Banana Pro.
  2. Use the same prompt and ratio for a controlled two-model test; note which provider the result actually used if available.

Compose for the intended shot

A beautiful square still may make a poor 9:16 opening frame. Specify subject placement, usable negative space, lighting direction, and the target ratio before rendering.

  1. Try “one cobalt mug on a table, right third, empty left side for title, soft window light”; choose the downstream video ratio.
  2. Generate and inspect whether the cup is fully visible and has room to move; revise placement rather than asking the video model to repair a bad crop.

Change one variable per revision

An uploaded or generated Reference image can help preserve a product’s shape while another model explores lighting. The connected @ picker controls which images are actually provided.

  1. Connect the approved cup still to a new Image node’s Reference image port and select its @ thumbnail.
  2. Change only the lighting to “rainy window reflection”; compare the handle, rim, and color against the source.

Use specialized presets as briefs

Specialized Image Tools provides character, scene, and spatial starting prompts; it still uses the Image generation node and an image model, rather than a separate spatial reconstruction service.

  1. Choose Image → Specialized Image Tools when a character turnaround or scene concept matches your reference need.
  2. Replace irrelevant preset details with your actual wardrobe, camera angle, and lighting, then keep only the approved result for video.

Recover from a model or reference failure

If a 302.AI image request fails, review the selected model, prompt, reference image and provider error before retrying. The Gemini route does not silently switch to a different model.

  1. If a selected reference is absent, check that the upstream image succeeded, the image port is connected, and its @ thumbnail is selected.
  2. If the model route fails, record the UI error and ask an administrator to check the relevant 302AI or TokenStar access; never paste credentials into the prompt.

助力大赛,创作继续!

+

今日 2,500 积分已到账。