The generate bar
Pick the model, set the frame, add references and see the cost before you run a shot.
The generate bar is where you set up a shot and run it. You pick the model, set the frame and length, add your references, and see the price before you spend anything. It opens at the bottom of the side panel, in place of the Light chat box, when you select one placeholder. Your messages with Light stay above it. A placeholder is an empty shot waiting to run. Generate Image, Generate Video and tools like Edit & Merge each add one to the board.
Light fills in the same bar when it places work for you. You can check or change anything before it runs.
The bar shows only the controls the chosen model supports. Switch models and the row changes with it.
The top row
The dark strip at the top names the job. It says Image, Video, or the name of the tool you started from.
Under the strip is a row of pickers. Say you are making a 9:16 reel of a silver ring. You would pick a video model, set the ratio to 9:16, choose 1080p and set the duration.
| Control | What it sets |
|---|---|
| Model | Which model runs. On a free plan, every video model and the premium image models show a lock and take you to pricing. See Models. |
| Aspect ratio | The frame, like 16:9 or 9:16. Hidden when an image sets the frame, as in image to video. |
| Resolution | Image size, like 1K or 2K, or video size, like 720p or 1080p. |
| Quality | Low to High on GPT Image 2, and up to Max on GPT Image 2.5 Flare and Sunburst. Higher takes longer and can cost more. |
| Quality or Fast | On Krea. Both cost the same, and Fast returns sooner. |
| Thinking level | On Nano Banana 2.1 and Krea 3. More thinking helps prompts with many parts. |
| Upscale factor | 2x, 3x, or 4x, on upscale models. |
| Duration | Clip length in seconds. Each model has its own list. |
| Sound | The speaker button. Turns sound on or off, on video models that can make their own audio. |
| Source audio | On video edits. Keeps or drops the sound from the clip you are editing. |
| Original sound | On motion control. Keeps or drops the sound from the performance clip. |
| Omni | On video models that take references, like Seedance and Happy Horse 1.0. Switches to reference mode, where you add several images, clips, and audio. |
More settings
The settings button at the end of the row holds the controls you change less often. It appears only when the model has one of them.
| Setting | What it does |
|---|---|
| Background | Auto, opaque, or transparent, on models that can output a cutout. |
| Web search | Lets the model look things up before it draws, for real products, places, or current events. |
| CFG | How closely the model sticks to your prompt, from 0 to 20. Higher is stricter, lower is looser. |
| Negative prompt | What to keep out of the shot, like "text, watermark, extra fingers". |
References
Below the row you write the prompt and fill the slots for anything the model should look at.
Reference images show the model a product, a face or a style. For a sneaker product shot, you would add a photo of the sneaker here. Each slot has Upload for a file from your computer and Canvas to pick a node already on the board. Models differ in how many reference images they take.
Video models that support them also have a first frame and an end frame. The clip starts on one and ends on the other. The swap button flips them.
In reference mode, Omni gives you separate slots for images, clips and audio. Drag them to reorder. Seedance 2.0 takes up to 3 reference audio clips, and Seedance 2.5 takes up to 10.
When you press Canvas, the board dims and says "Click any image or video on the canvas to use it". For an audio slot, it asks for any audio instead. Click a node and it fills the slot. Exit Selection Mode cancels.
Run
The run button shows the credit cost next to an arrow, before you press it. Change the model, resolution, duration or quality and the number updates. That number is the price. See Credits for what spends credits and what does not.
Press the run button. The bar shows Running… and the node shows Generating… while it works, and the result replaces the placeholder when it finishes. If the run fails, see When a run fails.