What’s on your mind?

Ask a question, explore an idea,
or create something new.

Idle

waiting for graph

Media jobs · Neo

Checking jobs…

Kyoko's graph, route decisions, tools, and sanitized thinking stream will appear here while she works.

Generation settings

Reasoning settings · Custom · up to 4,096 thinking tokens

Reasoning mode

Thinking token limit

Total response token limit

Auto lets the task choose whether to reason. On uses the Nemotron text model’s reasoning. Higher limits allow more work and can take longer; the model may finish early.

The total includes thinking and the final answer. Controls reserve at least 256 tokens for the answer when reasoning is enabled. Raising thinking may raise the total; lowering the total may lower thinking.

Applies to text responses. Image settings are independent. You can stop a response at any time.

Image settings · 1024 × 1024 · Auto steps (28–35)

Aspect ratioCustom ratio

1:4 to 4:1. Resolution adjusts only when needed to fit the supported range.

Output size

Generation steps

Step count

Auto estimates 28–35 steps from your prompt. Adjusting a control switches to manual. More steps do not always improve realism.

Preview frequency · Auto

Preview every N steps

Entering an interval selects a custom schedule. Auto uses four evenly spaced previews; Off shows only the final image.

The preview schedule follows the step count chosen for your prompt. Final image in 28–35 steps.

Each preview is a draft of the same image. More frequent previews add processing time; early drafts may look noisy. All captured versions remain available in the image’s Version menu.

Custom preview frequency is unavailable on this backend.

Final: 1024 × 1024 pixels. No cropping needed. Generated at the selected resolution.

Reset restores 1024 × 1024 with 32 steps. Maximum selects the largest size and 35 steps. Stop keeps the latest preview.

Video settings · 10 s · 512 × 288 · 35 steps

Size and aspect ratio

Video widthVideo height

Use multiples of 16. Larger videos take longer to generate. When animating an image, a different aspect ratio crops the source to fit.

Clip length and playback

Video duration in secondsFrame rate

10 seconds at 24 FPS. Higher frame rates give smoother motion and take longer to render.

Generation steps

Video step count

Preview frequency · Every 10 steps

Preview every N steps

Entering an interval selects a custom schedule. Auto uses four evenly spaced previews; Off shows only the final video.

35 generation steps per section. 3 draft previews per section, at steps 10, 20, 30. The complete video appears when all sections finish.

Each preview shows one draft frame from the clip. More frequent previews add processing time; early drafts may look noisy.

Custom preview frequency is unavailable on this backend.

Natural sound

Kyoko plans natural motion from your prompt and starting image. Previews show draft frames. Long videos can take tens of minutes to hours. Stop cancels the active job.

Sound is generated with the video. Include any spoken words in your prompt. Exact dialogue and lip synchronization are not reliable yet. Unmute the player to hear sound.