Cosmos 3 Super: lock this still before you buy motion (4cr)

Cosmos 3 Super is the stills row. Animate after approval. Do not generate the pack inside the video model.

Cosmos 3 Super is NVIDIA Cosmos 3 Super text-to-image. High-fidelity generation with optional agentic refinement and prompt expansion for strong prompt adherence. Content type is image. Category is text-to-image. requires_image is false. No max resolution listed. No quality chips. No durations. Audio is null. Aspects: 1:1, 16:9, 9:16, 4:3, 3:4. Display price: 4 credits. Features: Text-to-image, High-fidelity, Agentic, NVIDIA. The agentic pass is still a still. Prompt expansion is still a still.

NVIDIA high-fidelity, four credits, no clip

4 credits is the display price (flat, min 4, max 4). There is no 5s slider because Cosmos 3 Super does not move. High-fidelity is a plate claim. Agentic refinement is a plate claim: the model may expand and refine the prompt, then return an image. That loop is why people treat Super as "almost video." It is not. You still approve a frame.

The door is text-to-image. Term: text-to-image. Prompt expansion is the lever in the description — it improves adherence on the still, it does not add a camera. Ranked stills: best AI image generator. There is no NVIDIA provider hub on this site. Pick the slug.

Aspects are five: 1:1, 16:9, 9:16, 4:3, 3:4. There is no 21:9 and no 2:3 on this row. Pick the frame the video pass will inherit. A 4-credit 9:16 still you signed is a better I2V input than a 16:9 still you crop after motion.

No resolution ceiling is listed. Do not write 4K into the brief as a catalog fact. Inspect the file. Approve what you see.

Agentic refinement is still a still

Optional agentic refinement and prompt expansion are the differentiators versus a plain 2-credit creative stills row. They cost 4 credits and they still output an image. Do not skip the lock because the prompt got "smarter." The expanded prompt can still be a wrong pack. You catch that on the still, at 4 credits, not on a 10s generate.

NVIDIA in the feature list is the vendor, not a video family. Cosmos 3 Super's content type is image. Video is a later row you choose on purpose.

Prompt expansion is not motion

Strong prompt adherence is why you lock here when the brief is specific: product geometry, a listed setup, a camera that is actually a composition in a still. Image-to-video will inherit that composition. If you skip Cosmos and prompt the pack inside the video model, the video model will expand its prompt and move the result. That is two different expansions. Only one of them is cheap to reject.

The motion door is AI image-to-video. Use it after the 4-credit still is signed.

Approve, then animate

The test: would you send this frame as the key visual? If the agentic pass still missed the brief, rerun Cosmos. 4 credits. If you animate the miss, you bought seconds of high-fidelity wrong.

Do not generate the pack inside the video model. Cosmos 3 Super is the pack, as a still, with optional refinement. Motion is a different bill.

FAQ

Does Cosmos 3 Super generate video?

No. Text-to-image, 4 credits, no durations, no audio. Agentic refinement and prompt expansion still return a still. Animate after approval.

What does agentic refinement change?

It can expand and refine the prompt for stronger adherence. The output is still an image. It is not a clip, and it does not remove the need to lock the plate.

Do I need to upload a reference?

No. requires_image is false. The still you get at 4 credits is the file you later upload to image-to-video.

Which aspects are available?

1:1, 16:9, 9:16, 4:3, and 3:4. Match the cut you will animate. This row does not list 21:9.