Explore SenseNova U1

Provides the main natural-language instruction for the model. Include the subject, action, style, composition, and constraints that matter; clearer concrete details give the model more guidance about the intended result.

Provides the reference image for visual analysis or image editing. Leave it empty for text-to-image generation; when supplied, the prompt should clearly state what to inspect, preserve, or change.

Asks the model to perform a deliberate reasoning stage before producing the result. This can help with complex instructions and visual planning but uses additional time and tokens; disable it for simpler requests where speed matters more.

Lets the model rewrite a short prompt with extra visual detail before generating an image. This can improve composition from a simple request, but disable it when exact wording and strict prompt control matter.

Chooses whether the next token is sampled from a probability distribution or selected deterministically. Enable it for more varied responses that are affected by temperature and Top P; disable it when repeatability matters or when using beam search.

Sets the horizontal resolution of generated images or video frames in pixels. Larger values create more detail but require more memory and processing time; combine it with height to choose the aspect ratio.

Sets the vertical resolution of generated images or video frames in pixels. Larger values create more detail but require more memory and processing time; keep the height compatible with the model’s supported dimensions and desired aspect ratio.

Defines the approximate image area used when width and height are not supplied for an edit. The processor preserves the source aspect ratio while choosing dimensions near this budget; higher values retain more detail but use more memory.

Caps the total pixels processed from each source image to control memory and inference cost. Auto uses the model’s normal preprocessing choice; a numeric limit may downscale large images and reduce fine-detail recognition.

Allows the processor to scale source images down to the configured pixel limit before inference. Keep it enabled to control memory use; disable it only when preserving exact source dimensions is required and the image already fits the model limits.

Sets classifier-free guidance strength for the text prompt. Higher values prioritize literal prompt adherence but can oversaturate or reduce natural detail, while lower values allow more freedom.

Sets classifier-free guidance strength for the source image during editing. Higher values preserve more of the reference composition and appearance, while lower values permit the text instruction to make larger changes.

Chooses how classifier-free guidance values are normalized during image generation. Different modes constrain magnitude globally, per channel, or with CFG-Zero techniques, affecting stability and contrast; None applies no extra normalization.

Defines the normalized denoising progress at which classifier-free guidance starts. A value of 0 applies guidance from the first step; raising it leaves early composition more influenced by the model prior.

Defines the normalized denoising progress after which classifier-free guidance stops. A value of 1 keeps guidance active through the final step; lowering it lets late detail refinement proceed without guidance.

Moves the diffusion schedule so more denoising effort is spent in earlier or later noise levels. This can affect composition and fine detail, but the best value is model-specific and the default is the safest starting point.

Controls how many iterative denoising passes produce the image. More steps can refine detail and prompt adherence but increase generation time, and excessive values often provide little additional quality.

Controls how many items are evaluated in parallel during one forward pass. Larger batches can improve throughput but use more memory; reduce the value if processing runs out of device memory.

Initializes the model’s random-number generator. Reusing the same seed with identical model settings usually reproduces the same output, while changing it explores a different variation; exact reproducibility can still depend on hardware and implementation.

Caps how many tokens the model can add after the input. The model may stop earlier naturally, but a limit that is too low can cut off an answer; input tokens are not counted in this value.

Adjusts how strongly the model favors its most likely next token. Lower values produce more focused and repeatable output, while higher values allow less-likely choices and increase variety; it only has an effect when sampling is enabled.

Limits sampling to the smallest set of likely tokens whose probabilities add up to this value. Lower settings make output more focused, while a value near 1 keeps more alternatives available; it only affects sampled generation.

Limits sampling at each step to the specified number of most likely next tokens. Smaller values make output more conservative, while larger values preserve more variety; zero commonly means that this filter is disabled.

Changes the scores of tokens that already appeared in the generated text. Values above 1 discourage repeated words and loops, 1 leaves probabilities unchanged, and values below 1 make repetition more likely.