AI Features

AI Image Generation

View as Markdown

AI Image Generation creates product images, illustrations, backgrounds, storyboards, and other stills for an edit. Each result is added to My Assets as a new asset and does not overwrite its references.

What do I need before writing the prompt?

  • Using ChatCut’s image models requires an active Pro plan and enough ChatCut credits.
  • First decide whether to start from text, upload reference images, or match an existing project frame. Text-only generation needs no uploaded file.
  • To preserve a person, product, logo, or layout, prepare clear reference images. To match the edit, plan to use @ for a clip, time, or existing frame.
  • Decide the subject, composition, aspect ratio, visual style, number of images, whether the background should be transparent, and where the result belongs.
  • You can ask in natural language. If typing / shows an image-generation command for the current Agent, you can enter through that command instead. If no such command appears, write the prompt directly.

Prompts you can copy

Prompt

Generate a 16:9 product hero image of black wireless headphones on a light-gray stone surface. Use soft light from the left, reserve space for a headline on the right, and use realistic commercial photography. Do not generate text.

Prompt

Use @brand-screenshot as a reference and generate a 9:16 abstract background. Keep its blue-purple gradient and rounded geometric language, but do not copy any text or interface elements.

Prompt

Create three storyboard frames for the gap at @00:12: the same person using a laptop in a modern office, shown as a wide shot, medium shot, and hand close-up. Keep the clothing and environment consistent.

Generate an image with AI

  1. Optionally use + to attach a reference, or Selection Mode to add project context with @.
  2. Choose a ChatCut image model. In ChatCut Desktop, switch the AI composer to Image Gen and pick the model there. On the web, open + → Generation models, choose it under Image model, and describe the image to the Agent. GPT Image 2.6 Flare is the default; see Choose an image model.
  3. Set the aspect ratio, output size, image count, and Auto, Opaque, or Transparent background, either in Desktop’s Image Gen settings or in your prompt to the Agent. Transparent output uses PNG and works with GPT Image 2.5 or later and with Seedream 5.5.
  4. Enter the prompt. If an image-generation confirmation appears, review the request details and any credit estimate shown, then select Allow.
  5. Follow progress in the AI conversation or My Assets.
  6. Drag the finished image from My Assets to a video track. To revise it, reference the latest result and describe what should change and what must remain fixed.

ChatCut image models default to 1 image and accept 1–10 per request. Each image generated with a ChatCut model has a separate job, an asset in My Assets, and its own ChatCut credit charge.

Choose an image model

ModelUse it forOutput sizesReference images
GPT Image 2.6 Flare (default)Everyday images and drafts when speed matters; also GIF and SVG1K, 2K, 4KUp to 10
GPT Image 2.6 SunburstPrecise edits, dense text and layouts, and keeping a person or product exact; also GIF and SVG1K, 2K, 4KUp to 10
GPT Image 2.5 Flare, GPT Image 2.5 Sunburst, GPT Image 2Earlier GPT Image models1K, 2K, 4KUp to 10
Nano Banana ProThe model previously listed as Nano Banana1K, 2K, 4KUp to 14
Nano Banana 2.1Many references, or very wide or tall ratios from 1:8 to 8:11K, 2K, 4KUp to 14
Nano Banana 2The same ratios as Nano Banana 2.1, plus small 512 px output512, 1K, 2K, 4KUp to 14
Seedream 5.5PNG or JPEG output, and splitting an image into layers1K, 1.5K, 2KUp to 10

All of these require a paid plan. The editor or confirmation card shows the estimate for your settings before the job starts.

Make an animated GIF or an SVG

GPT Image 2.6 Flare and Sunburst can return an animated GIF or an SVG vector image instead of a PNG. Both are made at 1K and can use reference images, such as your own character, product, logo, or sketch.

  • GIF: a short loop of 2–16 frames, for a sticker, an animated icon, or a reaction. Describe the subject, the motion, and how the loop returns to its start. Add a transparent background when the GIF will sit over video. In ChatCut Desktop, Image Gen offers the frame count and frame timing.
  • SVG: a static vector image, best for logos, icons, and flat illustrations with few colors. Photographic detail does not translate well to vectors.

For animation you will keep editing, such as titles or lower thirds, use Motion Graphics. For real motion or anything longer than a few seconds, use AI Video Generation.

Split an image into layers with Seedream 5.5

Ask the Agent to split one PNG or JPEG image into layers with Seedream 5.5. The result is a base image plus separate transparent PNG layers that you can move, replace, or edit one by one. You can also ask it to recognize the text in the image. Layer splitting is done through the Agent; the Image Gen composer in Desktop does not have a layer control.

Prompt

Use Seedream 5.5 to split @poster into separate layers so I can move the product and replace the headline. Recognize the text too.

Add an image manually

If the image already exists, do not generate it again. Choose My Assets → Import → Files…, then drag it to a video track. Select the timeline item to adjust its position, scale, crop, and duration in the Inspector or canvas controls.

ChatCut credits and Codex usage

ChatCut’s Codex workflow uses Codex’s built-in image tool by default when it is available. Check Codex’s own image access and usage information for that route.

If you select a ChatCut image model, explicitly request ChatCut generation, or the workflow falls back to ChatCut because of tool availability or reference handling, generation requires an active Pro plan and uses ChatCut credits. Check the actual generation confirmation and Credits History for ChatCut jobs, even when the conversation stays in Codex.

See Generation Confirmations and Auto-Allow, Generation References, and Generation Credits.