subtitle_presetstring nullable
Caption style, spelled exactly as one of the model's subtitle_presets: any other name is quoted at the default style and refused at launch. Its subtitle_dynamic_presets animate word by word and bill at a higher rate, so pass the same preset to the estimate. Omit for the model's default style.
subtitle_colorstring nullable
Caption text colour as hex (#FFFFFF), for models whose capabilities report subtitle_primary_color.
subtitle_secondary_colorstring nullable
Colour of the words the model emphasises, as hex, for models whose capabilities report subtitle_secondary_color.
subtitle_alignment'top' | 'center' | 'bottom' nullable
Where the captions sit vertically, for models whose capabilities report subtitle_alignment.
base_model_idstring nullable
Base model to run, as an id (e.g. flux-dev), display name, or community alias.
modality'text' | 'image' | 'audio' | 'video' | 'three_d' | 'playable' | 'world'
negative_promptstring nullable
What to steer the output away from. Honored by the models whose inference-schema lists negative_prompt, and ignored by the rest.
widthinteger nullable
Output width in pixels. If not specified, a default resolution based on the model will be applied.
heightinteger nullable
Output height in pixels. If not specified, a default resolution based on the model will be applied.
auto_aspect_ratioboolean nullable
Derive the output's shape from the input instead of from width/height. Only on models whose capabilities include autoAspectRatio.
auto_resolutionstring nullable
The size tier of an auto_aspect_ratio run (e.g. '2K'), one of the model's autoResolutions. Unset on an image run, the smallest tier that holds the input image.
batch_sizeinteger
Number of outputs (1-16).
num_inference_stepsinteger nullable
Number of diffusion steps.
guidance_scalenumber nullable
prompt_strengthnumber nullable
How closely the generated audio follows the text prompt (0-1); higher means less variation. Honored by ElevenLabs Sound Effects and the Sonilo video-to-music models, and ignored by every other model. To control how closely an image or video generation follows an input image, set guidance_files[].weight instead.
quality'low' | 'medium' | 'high' nullable
Quality level: low, medium, or high.
prompt_enhancement'off' | 'fast' | 'standard' | 'quality'
How much effort the provider spends rewriting the prompt before it generates.
One ladder for a dial providers spell a dozen ways: an on/off flag (enable_prompt_expansion,
prompt_optimizer, enhance_prompt, expand_prompt, autoprompt, ...) or a mode (MiniMax's
disabled/fast/balanced/quality, Seedream's fast/standard, Qwen's none/quality). Member order is
the ladder. Each model declares the levels it offers as prompt_enhancement_levels, and the
provider mapping spells a level its own way; an on/off provider offers OFF and STANDARD.
This is the provider's own rewriting, distinct from Layer's prompt-enhance action, which
rewrites the prompt visibly before the request is made.
duration_secondsnumber nullable
Output length in seconds — a video clip's duration, or the clip length for audio models whose capabilities report duration_seconds_min and duration_seconds_max. A value outside a model's supported range is refused.
generate_audioboolean nullable
Generate audio with video.
keep_audioboolean nullable
Keep audio from input video.
draftboolean nullable
Render a quick, low-cost draft instead of the full-quality output, on models whose capabilities report draft. Complete the draft later with the model named by the generating model's draftCompletionModelId, passing the draft file as a DRAFT_VIDEO input (DRAFT_WORLD for a world, DRAFT_MESH for a mesh).
instrumentalboolean nullable
Guarantee a track with no vocals, on models whose capabilities report instrumental. Asking for one in the prompt is only a suggestion.
loopboolean nullable
Compose the clip to repeat with no audible seam, on models whose capabilities report loop.
output_format'mp3' | 'wav' | 'aac' | 'flac'
Container for a delivered audio file.
Providers name these their own way and Layer never re-encodes what one returns, so the
format a caller asks for is only keepable on models whose capabilities list it. Members are
named for their wire value: the frontend artifacts preserve enum names, and a name that
differed from the value would hand clients a format string the request API rejects.
lyricsstring nullable
Words to be sung, for models whose capabilities report lyrics. A separate input from the prompt, which describes the music itself.
fpsinteger nullable
FPS for LTX video generation (e.g. 25 or 50).
use_ta_poseboolean nullable
Deprecated: use pose_mode. Legacy rig-ready-pose toggle for 3D.
pose_mode'A_POSE' | 'T_POSE'
Canonical rest pose a character-mesh model is asked to generate in.
A_POSE places the arms angled down at roughly 45°; T_POSE holds them
straight out to the sides. Only meaningful for models that advertise the
pose_modes capability (e.g. Meshy V7).
Deliberately a light top-level module (like base_model_id), NOT under
pkg.models.inference: the Blueprint definition layer registers this as a
BlueprintType and migrates legacy node ports, and must do so without pulling
in the heavy pkg.models.inference package, which would perturb the Temporal
workflow-sandbox import graph and split pydantic class identity.
include_texturesboolean nullable
Include textures in 3D output.
detailed_geometryboolean nullable
Denser, more detailed geometry (Meshy V7 ultra mode). Costs extra; support varies by model.
remesh_geometryboolean nullable
Set false to skip the provider's remesh phase and keep its raw geometry. Support varies by model.
texture_resolution'RES_2K' | 'RES_4K' | 'RES_8K'
Edge length of a generated mesh's base color texture.
Offered where the model's texture_resolutions capability lists it (Meshy's and Tripo's own APIs).
remove_lightingboolean nullable
Strip baked highlights and shadows from the texture. Support varies by model.
mesh_density'ULTRA' | 'HIGH' | 'MEDIUM' | 'LOW'
How many faces a mesh keeps, as a level the provider sizes adaptively.
Stands in for a fixed face limit where the model's mesh_densities capability offers it.
preserve_uvsboolean nullable
Keep the mesh's own UV layout when retexturing. Support varies by model.
segment_by'AUTO' | 'PARTS'
What a mesh is split along.
AUTO finds the parts itself; PARTS cuts along the parts segment_part_names names.
Offered where the model's segment_by capability lists it.
segment_part_namesstring[] nullable
Parts to split along when segment_by is PARTS, at most 10.
quad_meshboolean nullable
pbr_materialsboolean nullable
generate_partsboolean nullable
face_limitinteger nullable
Face/polygon limit for 3D mesh.
enable_riggingboolean nullable
Rig the generated 3D mesh with a skeleton. Honored by the models whose capabilities report rigging.
enable_animationboolean nullable
Animate the generated 3D mesh with the motion of the animation reference set passed in reference_sets (category animation), or the provider's default motion without one; implies rigging. Honored by the models whose capabilities report animation.
use_speaker_boostboolean nullable
similarity_boostnumber nullable
Voice similarity boost (0-1).
style_exaggerationnumber nullable
Style exaggeration (0-1).
upscale_rationumber nullable
Upscale factor (e.g. 2.0, 4.0).
creativitynumber nullable
Creative variation strength for upscaling.
resemblancenumber nullable
Resemblance to original for upscaling.
denoisenumber nullable
Noise removal. On the models whose capabilities report denoise_range, in that range; a value outside it is refused.
compression_repairnumber nullable
Compression-artifact removal. On the models whose capabilities report compression_repair_range, in that range; a value outside it is refused.
halo_reductionnumber nullable
Removal of bright edge outlines. On the models whose capabilities report halo_reduction_range, in that range; a value outside it is refused.
recover_detailnumber nullable
How much of the source's own detail is kept. On the models whose capabilities report recover_detail_range, in that range; a value outside it is refused.
detailnumber nullable
How much fine detail the model adds. On the models whose capabilities report detail_range, in that range; a value outside it is refused.
grainnumber nullable
Film grain laid over the result. On the models whose capabilities report grain_range, in that range; a value outside it is refused.
refine_strengthnumber nullable
How strongly lettering and shapes are rebuilt. On the models whose capabilities report refine_strength_range, in that range; a value outside it is refused.
realismnumber nullable
Bias of added detail toward photorealism. On the models whose capabilities report realism_range, in that range; a value outside it is refused.
face_enhancementboolean nullable
Set false to skip the face-restoration pass, on models whose capabilities report face_enhancement.
face_enhancement_strengthnumber nullable
Strength of the face pass. On the models whose capabilities report face_enhancement_strength_range, in that range; a value outside it is refused.
face_enhancement_creativitynumber nullable
Facial detail the face pass invents. On the models whose capabilities report face_enhancement_creativity_range, in that range; a value outside it is refused.
target_region'all' | 'subject' | 'background'
color_preservationboolean nullable
Keep the source's palette, on models whose capabilities report color_preservation.
vectorizeboolean nullable
Vectorize the output image.
vectorization_mode'spline' | 'polygon'
vectorization_color_mode'color' | 'binary'
vectorization_layering'stacked' | 'cutout'
vectorization_filter_speckleinteger nullable
Drop specks smaller than this many pixels. Honored by the models whose capabilities report vectorization_controls.
vectorization_color_precisioninteger nullable
Color quantization level; lower merges more colors. Honored by the models whose capabilities report vectorization_controls.
vectorization_layer_differenceinteger nullable
Color difference that splits gradient layers. Honored by the models whose capabilities report vectorization_controls.
vectorization_corner_thresholdinteger nullable
Minimum angle, in degrees, kept as a sharp corner (spline mode). Honored by the models whose capabilities report vectorization_controls.
vectorization_length_thresholdnumber nullable
Longest curve segment before it is subdivided (spline mode). Honored by the models whose capabilities report vectorization_controls.
vectorization_splice_thresholdinteger nullable
Angle, in degrees, at which a curve is split (spline mode). Honored by the models whose capabilities report vectorization_controls.
vectorization_max_iterationsinteger nullable
Curve-fitting passes. Honored by the models whose capabilities report vectorization_controls.
vectorization_path_precisioninteger nullable
Decimal places kept in SVG path coordinates. Honored by the models whose capabilities report vectorization_controls.
transparencyboolean nullable
Put the output on a transparent background. Also accepted as remove_background.
reframeboolean nullable
Reframe/extend the image.
tileabilityboolean nullable
Generate a seamless texture that tiles without a visible edge. Honored by the models whose inference-schema lists tileability (list them with GET .../base-models?tileable=true), and ignored by the rest.
projection'EQUIRECTANGULAR'
The geometry an output image is drawn in, when it is not an ordinary flat picture.
projection_strengthnumber nullable
How strongly a model's 360 LoRA draws the projection, as its scale. Unset takes the LoRA's own.
refillboolean nullable
Outpaint/refill transparent areas.
nine_sliceboolean nullable
Deliver the image as a 9-slice UI element (a button, frame or panel that stretches): after everything else the run does, the plain band through its middle is found on each axis, the stretched regions are cleaned so nothing smears, and the insets are recorded as the file's nineSlice metadata. An image with no such band is delivered as drawn, without insets. Image outputs only.
expressionstring nullable
The expression a face edit gives the face, one of the names the model's capabilities list under expressions.
target_languagestring nullable
The language a dub speaks, as one of the ISO 639-1 codes the model's capabilities list under dub_languages.