SeamlessImageGenerationRequest class final
A text-to-image request with independently repeatable texture axes.
Both axes default to off. When either is enabled, the initial qualified
recipe is SDXS (512-DS) with its header-detected embedded tiny decoder,
no component overrides, 512 by 512 pixels, and CPU or Metal computation.
Load with flashAttention: false and vaeDirectConvolution: false.
Other families, decoders, sizes and runtime options throw
LlamaUnsupportedException before generation starts.
The runtime applies circular padding to both diffusion and decoder convolutions. Inspect a repeated tile: this improves continuity but does not promise that every prompt and seed produces a useful texture.
This separate type preserves existing implementations and subclasses of ImageGenerationRequest. Ordinary requests keep their original behavior.
- Inheritance
-
- Object
- ImageGenerationRequest
- SeamlessImageGenerationRequest
Constructors
- SeamlessImageGenerationRequest({required String prompt, String negativePrompt = '', int? width, int? height, int? steps, double? guidanceScale, int? seed, int count = 1, ImageGenerationSampler? sampler, ImageGenerationScheduler? scheduler, double? flowShift, bool circularX = false, bool circularY = false, ImageVaeTiling? vaeTiling})
-
Creates a request with independently optional circular axes.
const
Properties
- circularX → bool
-
Wraps left and right convolution boundaries.
final
- circularY → bool
-
Wraps top and bottom convolution boundaries.
final
- count → int
-
Number of images, from 1 to maxCount. Image
iuses seedseed + i.finalinherited - flowShift → double?
-
Timestep shift of flow-matching models, greater than 0 and at most
maxFlowShift.
nulluses the runtime's default for the model; other models ignore it.finalinherited - guidanceScale → double?
-
Classifier-free guidance scale from 0 to maxGuidanceScale.
nulluses defaultGuidanceScale; distilled models use 1, which also skips the negative prompt and halves the work per step.finalinherited - hashCode → int
-
The hash code for this object.
no setterinherited
- height → int?
-
Output height in pixels, with the same range, default and rounding as
width.
finalinherited
- negativePrompt → String
-
Negative prompt. Ignored when the guidance scale is 1, as distilled
models such as SDXS and SD-Turbo use.
finalinherited
- prompt → String
-
Text prompt.
finalinherited
- runtimeType → Type
-
A representation of the runtime type of the object.
no setterinherited
- sampler → ImageGenerationSampler?
-
Sampling method.
nulluses the runtime's default for the model.finalinherited - scheduler → ImageGenerationScheduler?
-
Noise schedule.
nulluses the runtime's default for the model and sampler.finalinherited - seed → int?
-
Random seed, 0 or greater.
nullpicks a random seed; the result reports the seed used either way.finalinherited - steps → int?
-
Sampling steps from 1 to maxSteps.
nulluses defaultSteps, which suits undistilled SD 1.x and 2.x models; distilled models such as SDXS, SD-Turbo or FLUX.1-schnell need 1 to 8.finalinherited - vaeTiling → ImageVaeTiling?
-
Optional decode tiling. Combining tiling with an enabled circular axis
is unqualified and throws LlamaUnsupportedException.
final
- width → int?
-
Output width in pixels: a multiple of 8 from minDimension to
maxDimension.
nulluses defaultDimension. Use the model's native size: 512 for SD 1.x and 2.x models such as SDXS and SD-Turbo, 1024 for SDXL and newer families. The runtime rounds SD 1.x and 2.x sizes up to a multiple of 64;GeneratedImage.widthreports the size produced.finalinherited
Methods
-
noSuchMethod(
Invocation invocation) → dynamic -
Invoked when a nonexistent method or property is accessed.
inherited
-
toString(
) → String -
A string representation of this object.
inherited
Operators
-
operator ==(
Object other) → bool -
The equality operator.
inherited